Shibani 25721005

TO STUDY THE IMPACT OF THE COVID-19 PANDEMIC ON
EDUCATIONAL PATTERNS AMONGST STUDENTS USING DEEP

LEARNING-BASED MECHANISMS
A
THESIS
SUBMITTED TO THE
SHRI JAGDISHPRASAD JHABARMAL TIBREWALA UNIVERSITY,
FOR THE DEGREE
OF
DOCTOR OF PHILOSOPHY
IN
Computer Science
BY
Mrs. Shibani Manoj Kulkarni
(Registration No 25721005)
UNDER THE GUIDANCE OF

Dr. Ajit Kumar
(Registration No JJT/2K9/SC/2576)
UNDER THE CO-GUIDANCE OF

Dr. Ranjit D. Patil
(Registration No JJT/2K9/SC/2304)
DEPARTMENT OF COMPUTER SCIENCE

SHRI JAGDISHPRASAD JHABARMAL TIBREWALA UNIVERSITY,
VIDYANAGARI, JHUNJHUNU, RAJASTHAN – 333001
Year 2023
i
DECLARATION BY THE CANDIDATE
I hereby declare that, this thesis entitled “TO STUDY THE IMPACT OF THE
COVID-19 PANDEMIC ON EDUCATIONAL PATTERNS AMONGST
STUDENTS USING DEEP LEARNING-BASED MECHANISMS” is my own
work conducted under the supervision of Dr. Ajit Kumar (JJT/2K9/SC/2576) and the
co-supervision of Dr. Ranjit D. Patil (JJT/2K9/SC/2304) approved by the Research
Degree Committee of the University and that I have put in more than 200 days / 600
hours of attendance with the supervisor.
I further declare that to the best of my knowledge, this thesis does not contain any part
of any work, which has been submitted for the award of any degree either by this
university or by any other university / deemed university without a prior citation.
(Signature of the candidate)

Name: Shibani Manoj Kulkarni
Registration No: 25721005
ii
CERTIFICATE OF THE SUPERVISOR
It is certified that, this work entitled “TO STUDY THE IMPACT OF THE
STUDENTS USING DEEP LEARNING-BASED MECHANISMS” is an original
research work done by Smt. Shibani Manoj Kulkarni under my supervision for the
degree of Doctor of Philosophy in Computer Science to be awarded by Shri
Jagdishprasad Jhabarmal Tibrewala University, Vidyanagari, Jhunjhunu, Rajasthan,
India and the candidate has put the attendance of more than 200 days / 600 hours with
me.
To the best of my knowledge and belief, this thesis

I. embodies the work of candidate himself / herself,
II. has duly been completed,
III. fulfils the requirements of the ordinance related to Ph.D. degree of University and
IV. it is up to the standard in respect to both content and language for being referred to
the examiner.
iii
CERTIFICATE OF THE CO-SUPERVISOR
It is certified that, this work entitled “TO STUDY THE IMPACT OF THE
STUDENTS USING DEEP LEARNING-BASED MECHANISMS” is an original
research work done by Smt. Shibani Manoj Kulkarni under my supervision for the
degree of Doctor of Philosophy in Computer Science to be awarded by Shri
Jagdishprasad Jhabarmal Tibrewala University, Vidyanagari, Jhunjhunu, Rajasthan,
India and the candidate has put the attendance of more than 200 days / 600 hours with
me.
To the best of my knowledge and belief, this thesis

I. embodies the work of candidate himself / herself,
II. has duly been completed,
III. fulfils the requirements of the ordinance related to Ph.D. degree of University and
IV. it is up to the standard in respect to both content and language for being referred to
the examiner.
iv
ACKNOWLEDGEMENT
Pursuing Ph.D. was truly a transformative experience for me. Throughout this
research journey, I was fortunate enough to get support and motivation from many
people. I would like to take this opportunity to express my sincere gratitude towards
all of them.
First, I express my sincere gratitude to my Ph.D. supervisor Dr. Ajit Kumar, Assistant
Professor, Shri Jagdishprasad Jhabarmal Tibrewala University Jhunjhunu, Rajasthan.
His guidance has been invaluable in the entire process.
I extend my deepest gratitude to my Ph.D. co-supervisor Dr. Ranjit D. Patil, Principal

and H.O.D (Computer Science), Dr. D. Y. Patil Arts, Commerce, and Science
College, Pimpri, Pune for his guidance and encouragement. His valuable insights have
been instrumental in shaping my academic and research progress.
I would like to thank all my colleagues and staff members of the department for their
co-operation. I thank the Lab assistant and non-teaching staff of the computer science
department for their technical support.
I am indeed thankful to the authorities and staff members from various colleges and
schools of Pune district and PCMC region for helping me in the process of student
data collection. I am indebt thankful to the students of the colleges and schools who
took time out and participated in the survey.
I extend my sincere thanks to all the authorities and staff members of Shri
Jagdishprasad Jhabarmal Tibrewala University Jhunjhunu, Rajasthan. It was a
pleasure working with JJT University and to avail the facilities provided by the
University.
I extend my heartfelt thanks to my husband and family members for their constant
support without which it was just impossible to achieve and complete this research
work. I thank my well-wishers and friends for their motivation.
Finally, I would like to dedicate all this work to my parents. Although, they are not
amongst us today, their blessings and well wishes would help me sustain throughout
my life.
v
ABSTRACT
Education is the cornerstone of all development and advancement, both personally

and socially. True learning brings about a positive change in the society. The Covid-
19 pandemic imposed longest closures of schools and colleges globally. It made the
world think about using technology in education to the maximum extent. Technology
had entered our lives before the pandemic and its usage was growing steadily.
However, enormous changes were observed due to the pandemic.
Many inequalities were raised and there was a need to address them. Gender
inequality, digital inequality, the gross enrollment ratio in urban and rural areas,
dropout rate of students, lack of nutrition, lack of medical help, need of social and
emotional support to the children and several problems came into picture. In this
adverse situation, the primary objective of education system was to carry on the
teaching-learning process with whatever infrastructure was available and make the
best possible effort to impart education.
In these years students’ promotion rate into higher classes was more, as an outcome
the passing percentage of students was high but the level of conceptual understanding
gained by them was questionable. Post pandemic all this had to be sorted out
efficiently by the teachers, students, parents and administrators. The pandemic made
the academicians and policy makers to redefine the education system. The factors of
personalized learning, preferential learning were brought about in this period. Thus,
there was a need to analyze first the effect that the pandemic has put on the children
and then reconsider the formation and policies of the academic model. There is a need
to integrate the technical aspects into the academic model. Another factor that the
pandemic has put forward is the need to encourage holistic development of the
students. This will help them survive in testing times of their life. The uncertain
situations that were brought by the lockdown had increased the level of stress,
anxiety, depression, financial instability, health issues among people.
The researcher has reviewed the earlier literature in education, psychology, and
current computer science trends to understand the changing approach of students
towards education. Thus, it was understood that the opinion of students about covid-
19 pandemic and its effect on their education and personality traits was something
that should be emphasized upon.
vi
Keeping this in mind the machine-learning and deep learning models were developed
to analyze the impact of the pandemic on their behavioral, motivational, physical, co-
curricular, and extracurricular aspects. There are several personality attributes, but
particularly these were considered to find the effect.
The research study is divided into three major modules. In the first part, four sub
modules were developed using the four machine-learning algorithms. The algorithms
used were KNN, SVM, NN, and RF. The second module used deep learning approach
and developed a CNN model. This was used to observe the emotions of the students
while they were filling the survey. The third module is the student survey application
tool. The tool is used to get student data on the five parameters mentioned above and
in parallel captures the video streaming. In the end, the tool displays the individual
effect on the student due to the pandemic. It also gives the overall effect on the
student. The results are displayed in graphical format.
The model developed will prove to be beneficial in analyzing the effect of the
pandemic on the students. This will help academicians, and educational authorities to
take mediating actions, which would help to improve the education. The results of the
research study can be useful to teachers, parents, students, academicians,
psychologists, in understanding the effect of the pandemic on the students and their
changing learning patterns.
vii
CONTENTS
Sr. No. Title Page No.

1 Table of Contents ix – xiii
2 List of Figures xiv – xvii
3 List of Tables xviii
4 Abbreviations xix – xxiii
5 Main Thesis Work Description 1 – 276
6 References xxiv – xlii
7 Appendix – Questionnaires xliii – xliv
viii
TABLE OF CONTENTS
Chapter / Topic Page No
CHAPTER 1 .................................................................................................................. 1
INTRODUCTION ...................................................................................................... 1
1.1 Introduction ...................................................................................................... 1
1.2 The outbreak of COVID-19 .............................................................................. 1
1.3 Impact of previous pandemics on Education .................................................... 2
1.4 Background of the study ................................................................................... 3
1.5 Evolution of Education ..................................................................................... 4
1.6 The Role of ICT in education ........................................................................... 7
1.7 Data Analytics Tools ...................................................................................... 12
1.8 Educational Schemes and government initiatives during COVID-19 ............ 13
1.9 Online Platforms that gained popularity during COVID................................ 15
1.10 Challenges in continuing Education: ............................................................ 19
1.11 Parent’s Role during Pandemic: ................................................................... 23
1.12 Teacher’s role and challenges: ..................................................................... 24
1.13 Challenges faced by students ........................................................................ 25
1.14 Problem at hand ............................................................................................ 30
1.15 Objectives of Research ................................................................................. 31
1.16 Hypothesis .................................................................................................... 31
1.17 Need of Study ............................................................................................... 32
1.18 Scope of Work .............................................................................................. 32
1.19 Thesis Outline ............................................................................................... 33
1.20 Summary....................................................................................................... 35
CHAPTER 2 ................................................................................................................ 36
LITERATURE REVIEW ......................................................................................... 36
2.1 Introduction .................................................................................................... 36
ix
2.2 Literature Review ........................................................................................... 36
2.3 Summary......................................................................................................... 83
Chapter 3 ...................................................................................................................... 85
Machine-learning algorithms and technology for model formation ........................ 85
3.1 Introduction .................................................................................................... 85
3.2 The technology used in the Model formation ................................................. 86
3.3 The Role of AI in Education........................................................................... 87
3.4 Significance of Machine-learning in education .............................................. 88
3.5 AI and ML approaches in education............................................................... 89
3.6 Need of Artificial Intelligence and Machine-Learning .................................. 93
3.7 Machine-Learning assistance to teachers ....................................................... 99
3.8 Machine-learning in Behavioral Analysis: ..................................................... 99
3.9 Machine-learning in Motivational Analysis ................................................. 101
3.10 Physical Analysis ........................................................................................ 102
3.11 Analyzing the co-curricular and extracurricular factors ............................. 103
3.12 The interdependency between the five parameters and the overall effect .. 104
3.13 Working of Machine-Learning: .................................................................. 104
3.14 Broad Categories of Machine-learning: ..................................................... 105
3.15 Data Pre-processing for the proposed model .............................................. 108
3.16 Machine-learning algorithms implemented in the proposed model ........... 111
3.17 Summary..................................................................................................... 127
CHAPTER 4 .............................................................................................................. 129

FACIAL EXPRESSION RECOGNITION MODEL ............................................. 129
4.1 Introduction .................................................................................................. 129
4.2 Computer Vision and Image Processing ...................................................... 129
4.3 Emotion Recognition .................................................................................... 131
4.4 Teacher’s perspective in understanding emotions of the student ................. 132
4.5 Different methods used in past few decades to understand emotions .......... 132
x
4.6 Facial Expression Methods........................................................................... 133
4.7 Facial action coding System ......................................................................... 133
4.8 Facial emotion Recognition Structure .......................................................... 133
4.9 Neural Network ............................................................................................ 138
4.10 Convolutional Neural Networks ................................................................. 139
4.11 Transfer Learning ....................................................................................... 142
4.12 Model Creation ........................................................................................... 142
4.13 Haar Cascade Algorithm ............................................................................ 153
4.14 Adaboosting ................................................................................................ 157
4.15 Weak classifier ........................................................................................... 158
4.16 Cascading.................................................................................................... 158
4.17 Applying the facial expression model on the live video streaming ............ 158
4.18 Workflow while implementing the trained model:..................................... 159
4.19 Summary..................................................................................................... 160
CHAPTER 5 .............................................................................................................. 161

THE STUDENT SURVEY APPLICATION ......................................................... 161
5.1 Introduction .................................................................................................. 161
5.2 Basics of a Web application ......................................................................... 162
5.3 Elements of framework selection ................................................................. 162
5.4 Framework Selection .................................................................................... 169
5.5 Django Framework ....................................................................................... 170
5.6 The Design of the Student Survey application ............................................. 171
5.7 Five major parameters of student’s personality considered for study .......... 173
5.8 Python Language .......................................................................................... 174
5.9 Python Libraries ........................................................................................... 176
5.10 Keras Application ....................................................................................... 182
5.11 The working of the Student Survey application ......................................... 183
xi
5.12 Summary..................................................................................................... 208
CHAPTER 6 .............................................................................................................. 210

RESULTS AND DISCUSSIONS .......................................................................... 210
6.1 Introduction .................................................................................................. 210
6.1 Evaluation of the ML Models for Physical, Motivational and Behavioral

analysis ............................................................................................................... 211
6.2 Performance Interpretation of the KNN model ............................................ 213
6.3 Interpretation of the SVM model.................................................................. 216
6.4 Interpreting the results achieved by the NN model ...................................... 219
6.5 Interpretation and Results of the RF model .................................................. 222
6.6 Mapping of Objectives vs Methods Implemented ........................................ 261
6.7 Potential use in online and distance learning (ODL) market: ...................... 262
6.7 Summary....................................................................................................... 265
CHAPTER 7 .............................................................................................................. 267

SUMMARY AND CONCLUSION ....................................................................... 267
7.1 Introduction .................................................................................................. 267
7.2 Summary....................................................................................................... 267
7.3 The interpretations of the Machine-learning Model ..................................... 269
7.4 Evaluation of the Physical, Motivational and Behavioral Model ................. 270
7.5 Major interpretations of the CNN model ...................................................... 273
7.6 The Survey application Results .................................................................... 274
7.7 The Jotform Survey Results: ........................................................................ 274
CONCLUSION ...................................................................................................... 277
7.8 Observations ................................................................................................. 277
7.9 Conclusions ............................................................................................. 277
7.10 Recommendations ................................................................................... 278
FUTURE SCOPE ................................................................................................... 279
LIMITATIONS ...................................................................................................... 279
xii
REFERENCES ......................................................................................................... xxiv
APPENDIX – Questionnaire .................................................................................... xliii
xiii
LIST OF FIGURES
Figure Page
Title
No. No.
1.1 Role of ICT Tools in Education 8
1.2 Online Education Access 21
1.3 Challenges of Students for Online Learning 29
1.4 Online Access in Rural Areas 30
3.1 Hierarchy of AI, ML and DL 86
3.2 Adaptive Logic Framework 90
3.3 Features of AI / ML 95
3.4 Schematic Representation of Machine-learning System 103
3.5 Machine-Learning Phases 105
3.6 Data Preprocessing of the Model 108
3.7 Implementing KNN Classifier in the model 113
3.8 Decision Tree from Bootstrapped Dataset 117
3.9 Decision Tree from Random Forest 118
3.10 Neural Network Working 120
3.11 Non Linear SVM 125
4.1 Image Processing – Blurred Image to Processed Image 129
4.2 Computer Vision and Machine-Learning 130
4.3 Face Detection Technique 133
4.4 Formula of Convolution Operation on the Image 140
4.5 CNN Model 144
4.6 Formula of Softmax Function 147
4.7 Explanation of parameters in the Softmax Function 147
4.8 Model Summary 149
4.9 Haar Features 152
4.10 Representation of facial features with Haar Features 153
4.11 Ideal Haar Feature’s Pixel Values 153
4.12 Realistic Sample Haar Feature’s Pixel Values 154
5.1 Structure of HTML 161
5.2 Structure of CSS 162
xiv
5.3 Coding Structure of JavaScript 163
5.4 Features of Ruby 164
5.5 Coding Structure of Python 165
5.6 Flow Chart of Student Survey and Results 170
Difference between Traditional Cycle and Python’s Development
5.7 173
Cycle
5.8 App Delivery Process using Python 174
5.9 Pickle Library Overview 176
5.10 NumPy Library Overview 178
5.11 Sci-Kit Learn Library Overview 179
5.12 Pandas Library Overview 179
5.13 Landing Login Page 181

5.14 Sign Up Page 182
5.15 Welcome Page of Survey 185
5.16 Survey: Personal Details 186
5.17 Survey: Physical Change Analysis 188
5.18 Survey: Motivational Change Analysis 189
5.19 Survey: Behavioral Change Analysis 190
5.20 Survey: Co-Curricular and Extracurricular Change Analysis 192
5.21 Survey: Submission Page 193
5.22 Survey: Result Link Page 193
5.23 Survey: Results in Text Form 194
5.24 Survey: Results in Graphical Form 195
5.25 Survey: Admin Login Page 196
5.26 Survey: Site Administration Home Page 197
5.27 Survey: Site Administration for Surveys 197
5.28 Survey: Site Administration for Users 198
5.29 Jotform: Personal Details 202
5.30 Jotform: Physical Change Analysis 202
5.31 Jotform: Motivational Change Analysis 203
5.32 Jotform: Behavioral Change Analysis 205
5.33 Jotform: Survey Submission 206
xv
6.1 Relation between Precision and Accuracy 211
6.2 Confusion Matrix for Behavioral Analysis using KNN 212
6.3 Confusion Matrix for Motivational Analysis using KNN 213
6.4 Confusion Matrix for Physical Analysis using KNN 214
6.5 Confusion Matrix for Behavioral Analysis using SVM 215
6.6 Confusion Matrix for Motivational Analysis using SVM 216
6.7 Confusion Matrix for Physical Analysis using SVM 217
6.8 Confusion Matrix for Behavioral Analysis using NN 218
6.9 Confusion Matrix for Motivational Analysis using NN 219
6.10 Confusion Matrix for Physical Analysis using NN 220
6.11 Confusion Matrix for Behavioral Analysis using RF 221
6.12 Confusion Matrix for Motivational Analysis using RF 222
6.13 Confusion Matrix for Physical Analysis using RF 223
6.14 Comparative Analysis of ML Models 230
6.15 Confusion Matrix for CNN Model 232
6.16 Model accuracy and Model Loss plots 233

6.17 App Result for Physical Effect 233
6.18 App Result for Motivational Effect 234
6.19 App Result for Behavioral Effect 235
6.20 App Result for Effect on Co-Curricular Activities 236
6.21 App Result for Effect on Extracurricular Activities 236
6.22 App Result for Effect on Average Emotions 237
6.23 Jotform Result for Effect on Behavioral Analysis 239
6.24 Jotform Result for Effect on Motivational Analysis 240
6.25 Jotform Result for Effect on Physical Analysis 242
6.26 Jotform Result for Effect on Co-Curricular Activities Analysis 243
6.27 Jotform Result for Effect on Extracurricular Activities Analysis 245
6.28 Jotform Result for Effect on Students’ Emotions 246
6.29 Analysis of Effect on Behavioral Analysis of Rural Students 247
6.30 Analysis of Effect on Behavioral Analysis of Urban Students 248
6.31 Analysis of Effect on Motivational Analysis of Rural Students 249
6.32 Analysis of Effect on Motivational Analysis of Urban Students 250
xvi
6.33 Analysis of Effect on Physical Analysis of Rural Students 251
6.34 Analysis of Effect on Physical Analysis of Urban Students 252
Analysis of Effect on Co-curricular Activities Analysis of Rural
6.35 253
Students
Analysis of Effect on Co-curricular Activities Analysis of Urban
6.36 254
Students
Analysis of Effect on Extracurricular Activities Analysis of Rural
6.37 255
Students
Analysis of Effect on Extracurricular Activities Analysis of Urban
6.38 256
Students
6.39 Analysis of Effect on Average Emotions of Rural Students 257
6.40 Analysis of Effect on Average Emotions of Urban Students 258
6.41 Online and Distance Learning (ODL) Market in India 260
xvii
LIST OF TABLES
Table Page
Table
No. No.
1.1 Universe and Sample for the proposed Study 33
1.2 Sample 33
3.1 AI / ML Tools used in Education 87
6.1 Classification Report for Behavioral Analysis using KNN 224
6.2 Classification Report for Motivational Analysis using KNN 224
6.3 Classification Report for Physical Analysis using KNN 225
6.4 Classification Report for Behavioral Analysis using SVM 225
6.5 Classification Report for Motivational using SVM 226
6.6 Classification Report for Physical using SVM 226
6.7 Classification Report for Behavioral using NN 227
6.8 Classification Report for Motivational using NN 227
6.9 Classification Report for Physical using NN 228
6.10 Classification Report for Behavioral Analysis using RF 228
6.11 Classification Report for Motivational Analysis using RF 229
6.12 Classification Report for Physical Analysis using RF 229
6.13 Classification Report for CNN Model 232
xviii
Abbreviations
Abbreviation Description
ML Machine-Learning
AI Artificial Intelligence
DL Deep Learning
ICT Information and Communications Technology
WHO World Health Organization
IIT Indian Institute of Technology
NIT National Institute of Technology
IIM Indian Institute of Management
LED Light Emitting Diode
PPT Power Point
PhET Physics Education Technology
PASS Personalized Adaptive Study Success
PM Prime Minister
TV Television
QR Quick Response
DTH Direct To Home
GSAT Geosynchronous Satellite
MOOC Massive Open Online Course
UGC University Grants Commission
INFLIBNET Information and Library Network
CEC Consortium for Educational Communication
IGNOU Indira Gandhi National Open University
CBSE Central Board of Secondary Education
NCERT National Council of Educational Research and Training
DAISY Digitally Accessible Information System
NEET National Eligibility cum Entrance Test
NIOS National Institute of Open Schooling
MS Microsoft
Mac Macintosh
iOS iPhone Operating System
Gmail Google Mail
CODEC Coder-decoder
PG Post-Graduation
xix
KG Kindergarten
3G Third Generation
4G Fourth Generation
ASER Annual Status of Education Report
UKG Upper Kindergarten
LKG Lower Kindergarten
NAS National Achievement Survey
PCMC Pimpri Chinchwad Municipal Corporation
JR Junior
SR Senior
SPPU Savitribai Phule Pune University
KNN K-Nearest Neighbor
SVM Support Vector Machine
NN Neural Network
RF Random Forest
PEBL Psychology Experiment Building Language
CNN Convolutional Neural Network
ILS Index of Learning Styles
BFO Beat Frequency Oscillator
BMI Body Mass Index
1D One Dimension
2D Two Dimension
3D Three Dimension
TREC Text Retrieval Conference
SPSS Statistical Package for Social Sciences
URL Uniform Resource Locator
MATLAB Matrix Laboratory
SWAYAM Study Webs of Active-Learning for Young Aspiring Minds
MLP Multilingual Learner Program
DASS Delhi Administration Subordinate Services
LMS Learning Management System
HOG Histogram of Oriented Gradients
HSCL Hopkins Symptom Check List
GSE General Self-Efficacy
ADFES-BIV Amsterdam dynamic facial expression Set-Bath Intensity Variations
xx
WSEFEP Warsaw Set of Emotional Facial Expression Pictures
MTV Making Thinking Visible
LBP Local Binary Patterns
ANN Artificial Neural Network
SDLC Software Development Life Cycle
PC Personal Computer
CCTV Closed Circuit Television
LBPH Local Binary Pattern Histogram
CART Center for Advanced Research and Technology
PANAS Positive and Negative Affect Schedules
BIU Behavioral Insights Unit
TAM Technology Acceptance Model
TPB Theory of Planned Behavior
PEOU Perceived Ease Of Use
PU Perceived Usefulness
PLS Performance Learning Systems
EEG Electroencephalogram
VGG Visual Geometry Group
JS JavaScript
AWS Amazon Web Services
DNN Deep Neural Network
EI Engagement Index
DB Database
FER Facial Expression Recognition
VADER Valence Aware Dictionary and sEntiment Reasoner
API Application Programming Interface
PISA Programme for International Student Assessment
PNAS Proceedings of the National Academy of Sciences
EFL English as a Foreign Language
LSTM Long Short-Term Memory
MCQ Multiple Choice Questions
ES Early School
JHS Junior High School
SHS Senior High School
GNA Global Network Academy
xxi
PPG Photoplethysmography
GSR Galvanic Skin Response
SSD Single Shot Detector
R-CNN Region-based Convolutional Neural Networks
PCA Principal Component Analysis
ReLu Rectified Linear Unit
JSON JavaScript Object Notation
OpenCV Open Source Computer Vision Library
CLI Command Line Interface
UI User Interface
HTML HyperText Markup Language
CSS Cascading Style Sheet
PHP Hypertext Preprocessor
PX Pixel
DOM Document Object Model
CPU Central Processing Unit
CRUD Create Read Update Delete
DRY Don’t Repeat Yourself
MVT Model-View-Template
UG Under Graduate
GUI Graphical User Interface
TPU Tensor Processing Unit
GPU Graphics Processing Unit
SciPy Scientific Python
NumPy Numerical Python
SKlearn Scikit-learn
BSD Berkeley Software Distribution
etc et cetera
MB Megabyte
GB Gigabyte
TB Terabyte
Email Electronic Mail
ID Identifier
PDF Probability Density Function
CSV Comma Separated Values
xxii
TP True Positive
FP False Positive
TN True Negative
FN False Negative
USD United States Dollar
INR Indian Rupees
ODL Online and Distance Learning
ERP Enterprise Resource Planning
IT Information Technology
SaaS Software as a Service
EdTech Educational Technology
FY Financial Year
xxiii
CHAPTER 1
INTRODUCTION
1.1 Introduction
The History of humankind has been witnessing several challenging situations such as
natural disasters, pandemics, wars and climatic changes. Such adversities bring in
gradual or drastic changes in the normal way of living of the human beings. However
being innovative in nature, this race always comes up with alternative solutions to
adapt to the new ways of living. The most recent experience that we went through was
the COVID-19 Pandemic. The Pandemic had a major impact on the people from all
lifestyles. The outbreak of this disease led to changes in the work-life of people in
both the public and the private sector. Owing to this situation, uncertainties started
evolving in each sector and education sector was not an exception.
Developing countries like India were on the path of acquiring technical resources and
skills in the field of education but were not fully equipped until the pandemic struck
us. The sudden occurrence of the pandemic and the lockdown imposed by it led to the
dramatic changes in education. The teachers and students had to move onto online
education overnight. This unexpected transition influenced various dimensions of the
students’ personalities.
Before the pandemic, a lot of research was carried out to find the influence of non-
academic factors on the academic performance of the students. In this research work,
many research papers were investigated to find the correlation of personality traits on
the progress of the students in academics. The researcher worked on finding the
COVID-19 pandemic’s impact on the overall behavior of the students through
machine-learning (ML) technique and associated models.
This chapter provides details on the COVID-19 pandemic, how the academic
institutions have evolved, support of learning platforms and ICT, changes and
challenges imposed in education; troubles and opportunities faced by the students and
similarly by the teachers during the pandemic.
1.2 The outbreak of COVID-19

This disease was first detected in Wuhan city in Hubei province of China in
December 2019. The first case of COVID-19 was recorded in India on 30 January
2020 in Thrissur district in the state of Kerala. There was a declaration from World
1
health organization (WHO) about this disease as pandemic on March 11, 2020. It was
observed that in human beings the infection could spread if someone comes in close
contact with the infected person through the respiratory droplets exuded by the
infected person while coughing, sneezing, or talking.
To control the outbreak of this highly contagious disease a nationwide lockdown was
announced in India on 24 March 2020. Initially the lockdown was declared for three
weeks but thereafter (with relaxations in certain domains it was extended until 31
May 2020. Later on, the Country experienced second wave of the pandemic towards
the end of February 2021, but nationwide lockdown was not observed at that stage.
The World Health Organization laid down some precautionary measures to prevent
the wide spread of this disease and those were imposed by the Indian Government;
Like
1. Regular washing of hands with soap and water
2. Use of Sanitizer
3. Wearing a mask
4. Social distancing
5. Avoiding crowded places
6. Keeping a minimum distance of one meter between people
The major implication was to avoid public gatherings, which was a very basic step
needed to break the chain of this disease. [1]
1.3 Impact of previous pandemics on Education

It is said, “History repeats itself”; almost all situations that we experience today had
already been experienced by previous generations in some manner. Thus, people
faced the situations that we went through the COVID-19 during pandemics like
plague, swine flu, malaria, cholera.
In the recent history, the outbreak of Spanish Flu pandemic in 1918 can be considered
as quite analogous to the existing situation of COVID. Earlier, there were not enough
medical equipment and testing kits. Thus, the detection and the treatment of the
disease thereafter was slower. The common practice that was followed to avoid
spreading of such diseases was to isolate the infected people. Apart from quarantine,
closure of educational institutions and public places were the prevention protocols
that were adopted to restrict the fanning out of the disease.
2
These practices were still relevant and continued amidst COVID-19. Usage of
innovative technology has proved to be a catalyst in transforming the educational
system in this pandemic. Though the schools, colleges and universities were closed,
we had an option of moving to online platform that was not prevalent during prior
pandemics.
Online learning turned out to be a positive change that the pandemic has reflected.
However, it could not replace traditional way of teaching as it has its own challenges
and limitations. Technology could be advantageous if used with a right approach.
The major challenge for the academicians and the researchers in the years to come is
to integrate it appropriately into the education system. There is a need to come up
with and educational framework having an apt balance between technology as well as
personal learning requirements of the students, which could even suffice teaching-
learning goals. [2]
1.4 Background of the study

Ancient India was blessed with a strong and structured education system. At that, time
academicians believed in the overall personality development of the students. The
curriculum and subjects were designed, to accomplish the all-round development of
the pupil. Efforts were made to imbibe self-discipline, punctuality, moral values,
ethics and practical skills in the students. The system also focused on the development
of inner- self, which could lead to self-realization and would help the student turn into
a better human being.
Through these historical evidences in the literature, one can clearly say that the
personality factors definitely played a crucial role on the overall academic
performance of the students. These factors were given importance while training the
students as it helped the teachers to identify the strengths and weakness of the
children. The teachers were able to find the areas of interest and individual expertise
of the child through the insight gained into the student’s personality. Thus, more
confident and successful individuals were developed in the society.
As education evolved value education, physical education were again included in the
system. From traditional education until date, we cannot deny the effect of personality
factors on education. Human and child psychology also explains the importance of
psychological factors on the academic growth of the children. The Psychological
3
literature details the relation of learning styles, learning strategies, motivation, and
behavior on the academic achievement of the students.
The big five factors that were recognized in the psychological study always had high
correlation with the academic performance. Psychometric tests and counselling
sessions help the students to decide their career path. In this phase of COVID
pandemic there is a need to find the influence of the disease and the lockdown on the
behavior and motivation of the students and their approach towards learning. The
research undertaken tries to uncover the effect of COVID pandemic on the learning
patterns and overall behavior of the students through a deep learning model.
1.5 Evolution of Education

Since olden days, India was well known for its rich and varied heritage. Travelers
from all over the world used to visit India to experience the culture and Educational
system. Education originated in India through Rigveda. During those days, teaching-
learning process was conducted in an open environment. People were close to nature.
In the Gurukul system formal education of the children started at the age of eight and
they were supposed to go to the guru’s place for acquiring lessons.
The Gurus focused on capacity building of the students and inculcated life skills in
them. Guru-Shishya enjoyed a strong affectionate bond with each other. Gurus
enforced teaching learning that could enhance the cognitive abilities of students and
groomed them in spiritual and physical aspects. The educational evolution on three
major factors is explained below.
1.5.1 Equality in education

In India, education is of utmost importance since ages. Education is a systematic
process that is essential to transform human beings into social beings. An educated
person turns out to be a civilized person with the acquisition of various skills,
discipline, values, and knowledge.
India had always been known for its glory and it was reflected in its structured
education system. Education originated in India around 1500 B.C. In those days,
education was available for all regardless of caste, creed, religion or gender.
Women enjoyed freedom. Gargi, Maitreyi were the famous female scholars of ancient
India. There were basically two categories firstly Brahmavadini’s who dedicated their
whole life seeking knowledge and studied Vedas and the second category belonged to
4
those who could acquire the basic reading and writing skills and got married. They
did not study the Vedas.
This status of women education in ancient India is quite contrary to the low literacy
rate of females in India after independence. Efforts are being made by the
Government to improve this situation. In the 21st century, condition is changing as we
could find women taking up key roles in almost every sector of the society.
Later, in the Vedic period social divisions appeared in the society and education was
imparted to a privileged class depending upon their caste. Women, labor class, and
tribal got a backseat and their education was neglected. This scenario continued
during the medieval period when patriarchal society came into existence. Even then,
slaves, women and tribal were deliberately kept deprived of education.
This condition did not improve even during the British rule, though many schools,
colleges were established during this era. In existing times lot of Government,
initiatives are taken up for woman empowerment and their education. In addition,
there are many government schemes to initiate and uplift the underprivileged class
trying to turn them into employable skilled-force and measures are taken up for the
education of their children.
1.5.2 Quality in education

Religious preaching, value education, overall personality development were included
in the Indian education system. The society became more complex day-by-day and
was split based on caste during medieval period. Brahmins were made proficient in
teaching skills; Kshatriyas acquired expertise in war skills, Vaishya were trained in
trade and commerce while Shudra were skilled for ground level working. During the
medieval period, architecture, paintings, crafting skills also gained importance.
Modern education emerged during the British rule. The value of craftsmen and
artisans declined during the British regime. Professional courses like Engineering,
Medicine and Law got importance. Initially, British officials were not much
concerned about the education of Indians.
Slowly, they realized that they would require skilled and educated people to assist
them. They were also in need of people who were proficient in native as well as
English Language.
In 1813, Charter Act was passed to initiate education of Indians. There were some
implementation problems as there was controversy regarding the instruction medium
5
and style of education. This was resolved in 1835, by the British Governor General in
India, Thomas Macaulay, when he came up with Macaulay’s Minute. This was an
education act, which imposed primary instruction medium as English.
The 1854 Wood Dispatch act encouraged Women education. Primary education
should be in vernacular language and English should be at the University level were
the major recommendations of the act. In 1882, Hunter’s Commission made
improvements in primary education. Student’s preparation for University entrance
exam and imparting professional education were the major suggestions of this act.
Dr. Sarvapalli Radhakrishnan provided some valuable inputs, which molded the
academic system in a better way in the post-independence period. He proposed that
before University education there should be 12 years of pre-University study. He
advised three basic objectives of education those were General Education, Value
Education and Professional Education.
1.5.3 Infrastructure in Education

Initially lessons were imparted to the students in an open environment. Later on with
the establishment of the famous Universities like Nalanda and Takshashila introduced
the concept of infrastructure was introduced.
India became a famous education hub. Scholars from all over the world visited this
University to gain Knowledge and skills. Chandragupta Maurya – the founder of
Maurya Empire, Panini – the Grammarian, Charakh – the father of Medicine and
Chanakya – the economist were the great scholars who studied in this University.
Around sixty-seventy courses were available which included subjects like – Arts,
Medicine, Law, economics, Vedic literature, Performing arts, archery, Mathematics,
astrology. Students were even trained for debates and group–discussions. Nalanda
University was known for its biggest library. Students were supposed to clear
entrance exam to take admission.
Meta-Physics, Literature and various other subjects were taught. A Chinese scholar
XuangZang visited Nalanda and took admission to pursue Yoga. Vallabhi,
Vikramshila, were some of the other famous Universities of those times. In medieval
India, Islamic education flourished. In British India many Schools, Colleges,
Universities, and district schools were established.
Today there are many prestigious institutions such as IITs, NITs, and IIMs in India.
Still there are students who apply in foreign Universities for education. According to a
6
report of Parliamentary panel on education 40% of the government schools do not
have either electricity or playgrounds.
There are certain drawbacks in funding and utilization of funds resulting in serious
infrastructural problems (3). India is still a developing country. Especially in rural
India, there is a need to take effective measures to set up a strong infrastructure,
which could facilitate better learning experiences to the student community in the
years to come and could sustain in pandemics and disasters. [11]
1.6 The Role of ICT in education

With the advent of Computers and the emergence of information, communication
technology (ICT), there were gradual changes made in the educational sector. The
dynamic platform provided by ICT tools helped to represent academic content in a
well-organized manner. The basic objective of ICT-assisted teaching learning is to
increase students’ accessibility to the content and upgrade the overall quality of
education. ICT devices help users and content providers to interact with each other.
ICT, which became a buzzword today, really requires careful attention by teachers.
ICT could be a hardware, software, process or a system that helps to create, retrieve,
store, manipulate, send and receive digital information.
It is high time that educators find better way to integrate ICT in teaching learning for
betterment of education in general and students in particular. The major concern of
the teaching fraternity is to explore the suitable pedagogical approach that would
match the content to be taught. ICT enhances understanding of the concepts, brings
effectiveness in teaching, and makes learning interesting.
There are various parameters such as strength of the class, age, demography, subject,
cognitive abilities of students, physiological health and social environment that play a
vital role in ICT tool selection and integration. Apart from this reliability,
accessibility, availability of the tool also needs to be considered while integrating the
tool with academic platform.
ICT opened various options such as videos, mind mapping, graphs, smart classrooms,
charts, and simulations that helped in gaining fruitful learning outcomes. ICT also
promotes education through radio stations and television programs. [10]
7
Fig. 1.1 Role of ICT Tools in Education
1.6.1 Need of ICT in Education

The students have become smarter and are expose to technology that they use for their
daily transactions. Teachers could no longer stick to the chalkboard method when
everything around the children is changing. Though prior to pandemic ICT was
accommodated in schools and colleges.
The COVID pandemic boosted the use of ICT in education. ICT imbibes creativity,
collaboration, and engagement amongst students. It provides personalized approaches
and reveals creativity of learners. ICT helps teachers extensively in lesson planning,
preparation, administration, assessment and evaluation. Thus saves a lot of time of
teachers.
It is an effective tool for delivering content. Teachers could then focus on student’s
individual learning styles and preferences. Thus, ICT when used appropriately helps
attainments in all subjects. Embedding ICT in education is a stepwise process wherein
the teachers should be made familiar with the technology first so that they could
utilize it, integrate it, reorient and finally it would evolve. Hence, we can say that it
helps in capacity enhancement of the teachers.
1.6.2 ICT Tools for education
1. ICT tools used in teaching

 Computer/Laptops
The credit of computer’s invention goes to Charles Babbage who intended to have a
device that does complex calculations. However, today it is getting used for a plethora
of purposes like office work, online education, engineering and commercial
8
applications, playing games, internet access, sharing data, entertainment, visual
creativity and so on. They work on different operating systems like Windows, UNIX,
and Linux.
We are now into 5th generation of computers that are capable of managing Artificial
Intelligence or commonly known as AI. Laptop is a version of computer that does
exactly the same job a normal desktop would do, however with lot of flexibility.
People can carry it anywhere and as it is battery driven, it can run without a power
cable for quite some time. Due to increase in its demand, they have become powerful
yet quite inexpensive compared to the generations before. They have become a
necessity nowadays and are found almost in all homes.
 Projector
A projector is an optical device to project the content seen on a monitor of a computer
on a bigger screen typically a white screen or a wall. There are two types of a
projection – Front and Rear.
Front projection is typically used in a closed room setup and where we do not have
much space behind the projection screen. While, a Rear projection is generally used in
open spaces and for commercial events. With the invention of LED screens, the
demand of projectors is going down and so the cost too.
 Presentation tools
The purpose of the presentation tool is to assist teachers in compiling the subject
content in a precise and organized manner. Power point (PPT), Google slides, Prezi
are some of the tools used frequently by the faculties. Every tool provides unique
features. There are many slide templates in power point and other features, which
could be used for content presentation.
 Interactive white boards

Interactive white boards are also called as “Electronic Whiteboards”. They help in
making classroom lessons interactive. The teaching content could be shared on this
board using a digital projector.
These interactive boards provide student engagement, which is the crucial part for
most of the educational institutes. In addition, it enhances collaborative learning
through group discussion and brainstorming techniques. Moreover, students can relate
the use of technology and the lessons learnt.
9
2. ICT Tools used for education requiring internet facility
a. YouTube
YouTube is an application that allows common people to watch videos that are posted
from different sources. YouTube is source of education, entertainment and marketing
videos. In today’s world, even the common people can create videos and post it online
on YouTube. Steve Chen, Chad Hurley and Jawed Karim who were prior employees
of famous American company PayPal had registered YouTube in 2005.
b. Blogs
Learning could happen in a better way when learner could reflect the activity or
experience. Blogs provide the platform to share the understanding of a particular
content. Also in the process of learning, there could be certain difficulties or
confusions, which could be resolved through blog writing as we could generate
community of learners and could know the perspectives of others on the same content.
Thus, it helps in content articulation and understanding. Blog writing should be
encouraged among student community, as it is an effective reflective tool for students.
Blogs will be more helpful if teachers promote the comment section usage so that the
learners will be motivated by peer comments.
c. Wiki
Wiki is an online free content encyclopedia that allows people from different parts of
the world to connect with each other. Referring to Wiki pages, they can share and
access available knowledge. People are also encouraged to share their expertise areas
and knowledge on this online platform so that other needy people can refer and learn
from it. Largely all these wiki platforms are user friendly, easy to browse and the
content on this platform is also easy to search.
d. Google Map
Google Map is an application to search and pinpoint the geographic locations.
Probably, it is the most widely used application all over the world by almost
everyone. It is used to show the directions from one location to other and thus, people
can easily get to the desired places without much of difficulty.
It is integrated in all advanced vehicles and even in the taxis, which are used by cab
drivers to locate the places for the passengers. It can be used as a live location
10
tracking for people. In education, it can be used to explain about geographical
location of cities, rivers, historical places and monuments.
e. Google docs
Google Docs is an online tool that allows creating spreadsheets and documents on the
cloud. It also allows editing and storing these docs online. As the documents are
stored on the cloud and are accessed through internet, these can be accessed from
anywhere. Google provides certain free space to store the data, thus even fixed
computer or laptop is not necessary to work with the documents.
f. Google
Google is one of the earliest search engines. It collects the information from entire
world, stores it on the servers, indexes it and provides that information to the users on
demand and on the fly by crawling through various servers. Today with the
advancement of technology, internet has reached to every home and each person and
Google search engine has become an integral part of each one of us. Although, there
are various other search engines, 92% of world’s information is searched through
Google.
3. ICT learning tools for educational games

a. Flash Games
b. Serious Games
c. Mainstream Games
4. ICT tools for specific subjects like Mathematics and Science
Geo-Gebra is an application that could be used for improving the mathematical
visualization. Similarly, PhET, an interactive simulation tool that helps in better
understanding of scientific experiments. This simulation platform provides real-time
experience to the children when there is lack of laboratory infrastructure.
Most of the schools at an elementary level do not have enough laboratory facilities
thus such virtual platforms help students to understand the practical aspect of science
subject.
1.6.3 Advantages of ICT

1. Easily accessible
2. Provides Remote Access
11
3. Incorporates learner centric approach of teaching
4. Saves time
5. It is useful for differently abled children and individuals
1.6.4 Disadvantages of ICT Tools in education

1. There is so much information available on the internet that students sometimes end
up in confused state of mind. They cannot figure out what to study and in how much
detail a particular topic should be covered.
2. Students have lost the habit of taking self-notes which is very important for college
students.
3. Data security and privacy is also an important issue that needs attention.
4. Students sometimes may be involve in playing games or could access unwanted
data or videos. Thus, we need to protect students from these distractions.
5. Online learning through ICT for long hours is not good for health especially for
eyes and posture of children.
1.7 Data Analytics Tools
1.7.1 The open learning initiative

The tool helps teachers to upload lecture notes, which could be utilized by the
students. Individual feedback could be given to the student. The teacher is able to
monitor the progress of the students through a dashboard and could extend support if
the students are facing any practical difficulties.
1.7.2 Personalized Adaptive study success

It is a powerful tool, which is helpful to find the involvement of students in online
learning.
Personalized Adaptive Study Success (PASS) is a very effective tool that provides an
indication to identify the engagement level and willingness to continue the online
learning platform for the beginners. This is identified on the basis of multiple factors
like personal nature and preferences, social profile, inputs from academic records and
physical content derived from various sources like individual study circle, friends,
study platforms and forums.
The engine supported by data science and analytics then processes this content. This
analysis is used by PASS to come up with the graphical indicators, level of
12
achievement, assessment done for self, and points towards social discussion, advices
content & actions and provides a customized environment for each individual.
Teachers to pinpoint the students who are getting challenged in the online learning
and potential withdrawal from it.
1.8 Educational Schemes and government initiatives during COVID-19

There are many challenges and opportunities raised in the educational sector due to
the pandemic. The Indian government came up with several initiatives to ensure that
the students at all levels receive equal access and resources to continue education
during the pandemic irrespective of their location, financial background, age, gender
and other factors that directly or indirectly create hindrance in their education. Indian
government came up with certain initiatives, which ensured learning for all students
including the remote locations.
A comprehensive initiative named “PM e-VIDYA” was launched. The initiative
provides online, digital, on-air education and various modules were introduced under
this, which are enlisted below:
Objectives of PM e-VIDYA: The basic motive of this scheme was to provide
education to all the students and minimize the effect of pandemic on education.
Various educational models were launched under this PM e-VIDYA to impart quality
education. School and college education are very crucial years of the students-life for
their academic growth. Keeping this view, the scheme was launched. Through these
modules online education was provided so that the physical presence could be
avoided which was needed to control the spread of the disease.
1.8.1 Important features of PM e-VIDYA:

1. The scheme facilitated education to more than 25 crore schools.
2. Around 100 top ranked Universities started online education for students after 30
May 2020.
3. Swayam Prabha TV channel was launched for those students who were not having
internet facility.
4. E-content and QR code energized books were provided for all the classes through
Diksha platform.
5. Radio podcast facility was provided for visually and hearing impaired students.
13
6. Competitive exam coaching through online mode was also provided through this
platform.
Models under PM e-VIDYA

1. Diksha: Digital infrastructure for knowledge sharing this program was introduced
with the idea of “One-Nation –One Digital Platform” which could provide quality e-
content to the students. This portal provides learning material according to the
prescribed syllabus of the schools. It provides content in English as well as many
other languages. The platform also contains assignments and activities related to the
lessons, which could help students to summarize and comprehend the lesson concepts.
[12]
2. Swayam: This portal offers free and best learning resources to students. The
program includes video lectures, self -analyzing tests, special notes and material on
relevant subject topics and has a discussion forum where the students can raise
queries, doubts and could get those resolved. [16]
2. Swayam-Prabha: “One class-One channel” where one channel for each class from
class 1 to class 12 was initiated. Swayam-Prabha has 34 DTH channels that deliver
educational programs, which are available 24*7. Every day a new content is created
that is streamed for four hours and then it is repeated five times so that the students
could avail it according to their convenient time.
This academic content is telecasted through GSAT-15 satellite. The channel covers
higher education programs under arts, performing arts, commerce, science, medicine,
engineering, law, humanities, agriculture and many more. These certificate courses
are designed under MOOC courses.
Special courses were developed for schoolchildren of class 9th to 12th standard with
the motive to clear their conceptual understanding. Modules were also designed for
helping these students prepare for entrance exams for seeking admission in
professional courses after 12th standard. These modules also help students prepare
better for competitive exams. IIT’s, UGC, CEC, IGNOU contribute in content
preparation. The web portal is managed by INFLIBNET.
CBSE Podcast- Shiksha Vani, Mukta Vidya Vani and radio podcast ensures the use of
community radio for those who are visually impaired or does not have enough
facilities could listen to these educational audio.
14
Shiksha Vani is specially designed for CBSE students from ninth to 12th grade in line
with the NCERT syllabus. The study material was written in very easy and
comprehensible language
Mukta Vidya Vani programs are beneficial for the students of secondary, higher
secondary and vocational courses. This program has proved to be the largest open
schooling system around the globe. It facilitates an interactive module thus
encourages two-way communication.
Apart from these Doordarshan and All India radio are streaming educational content
through radio, Television and YouTube. These classes aim to cover the syllabus of
middle school and high school students, also help board students to prepare for
competitive exams. Previous years papers are available which these students can take
as practice tests. Stories and quizzes are also included in this module.
Competitive exam preparation: The Higher education has created Abhyas mobile app
through which video lectures were offered in the subjects of physics, chemistry,
mathematics and biology.
The NEET department maintained the module. Students get one test per day on the
mobile application, which would help them prepare better.
DAISY: Educational content for differently abled students was developed in sign
language on (DAISY) Digitally Accessible Information System and it was displayed
on NIOS website or YouTube.
Following are the benefits of “PM e-VIDYA”:
1. Students will get online education.
2. Special radio podcast facility will be provided for visually and hearing impaired
students.
3. E-content will be provided for visually and hearing impaired students.
Manodarpan: This initiative was put forward to provide psychosocial support to
students, teachers and families for mental and emotional well-being in this global
pandemic.
The support was provided by website, toll-free helpline-number, interactive chat
platform, national directory of counselors. Government has made an effort to expand
e-learning framework by including open, online and distance education. [3][4]
1.9 Online Platforms that gained popularity during COVID
15
1.9.1 Microsoft Teams:
It is an online communication platform established in 2017 and was extensively used
during the pandemic. It turned out to be a digital workspace as people from most of
the working sectors had to switch to work from home set-up.
Important features of Microsoft Teams:

1. Secured Communication: Microsoft teams facilitate secured / authentic
communication between students within schools, colleges, and universities.
Educational institutes of different districts, regions could also be connected by setting
up a call or scheduling a meeting. Teams also offer chat option including administered
chat for students.
2. Learner Engagement: Interesting content could be developed with the help of
different applications and websites thus students could avail summarized study
material. Assignments, activities could be shared with the students, which they could
solve. Thus, it promotes personalized learning experiences and activity based
learning. Meetings could be scheduled for conducting online classes.
3. Tools for assessment: Teams have the form option through which teachers can
create quizzes. Grades are given automatically once the quiz is solved and submitted.
This saves time of the teachers, which could be better utilized in some other
demanding task related to teaching-learning process. Teachers can design digital
grade books, which maintain records related to assignments, unit assessments, class
test, and activities. Individual progress of student could be tracked.
4. Collaborative platform: Microsoft Teams provide a collaborative platform so that
user could collaborate in live sessions on excel MS word and power point. Most of
the learning Management tools can also be connected to teams, which encourages
group discussions, and teamwork that is needed for better understanding of certain
subjects. The collaborative platform is beneficial for practical subjects and project
work.
5. Independent of the location: There is no binding with respect to the place or region
where the student belongs. In addition, an inclusive and classroom experience is
provided through this platform. Recording option helps students to revisit the sessions
and learn certain concepts that they might have missed out or not understood.
6. Diagnostic assessment: Individual progress could be monitored and students at high
risk of failure as well as dropout rate could be intervened at an early stage. In addition
16
to this teams offer some tools that help in creating learning goals for a particular
academic year and help in identifying subject that require more attention. [5]
1.9.2 Zoom
Considering the COVID-19 pandemic, the use of video conferencing applications
became normal. These tools helped for establishing face-to-face communication in
both professional and personal contexts. The financial sector, government sector,
software companies, and religious communities, tend to use zoom as a video
communication tool. Not only had this but people wishing to talk with their family
and friends also preferred this platform.
One can use several collaborative features offered by zoom like audio and video
conferencing, webinars, and live chats. The convenient part about zoom is that there is
no need to create an account for using it. In addition, the software works with Linux,
Windows, Mac, iOS, and Android devices hence anyone could use it easily.
We can record meetings that could be started manually or automatically. Screen
sharing and polls could also be utilized effectively. It also has the provision for
whiteboard and encrypted calls. The chat option in zoom meetings could be used to
discreetly message an individual or a group of people during a meeting.
Breakout rooms help in dividing a single call into various groups, allowing
participants to carry on their respective talks. In the academic world, psychology or
history teachers predominantly used these breakout rooms. For example, if the teacher
has to conduct the role-play activity online then first the theme would be explained
then the roles would be assigned to the individual and then they could split into
groups and prepare for their respective roles. Similarly, to practice drama and do
rehearsals online the breakout rooms proved to be beneficial to the performing art
teachers and students.
Virtual backgrounds, one of the platform’s more interesting features, allows the user
to use an image or a video in the background.
Raise hand option: This is a virtual feature that enables the user to virtually raise a
hand. It notifies the speaker or the organizing committee members that the participant
wants to add some points to the ongoing session or wants to ask questions.
There are certain control features like the waiting room and mute participants, which
were helpful in online class control. Students who tried to join late could be kept on
17
hold in the waiting room. In addition, to avoid unnecessary disturbances the teacher
could mute the student participants.
A maximum of 1000 participants can join in a zoom meeting. The free version of
zoom allows users to schedule a meeting for a maximum of forty minutes. This is the
limitation of zoom. [17]
1.9.3 Google meet

One of the most accessible video conferencing tools nowadays is Google Meet. It
became popular since the pandemic. It is necessary to have a google account to create
and host google meet sessions. Once we have an account, we may use the Google
Workspace toolbar to start or arrange a meeting.
The host schedules the meeting and the link is provided to the participants. To attend
a meeting through a web browser one does not need a google account. To attend the
meeting through mobile devices users must download the google meet application and
log in to a Google or Gmail account. Google meet’s free plan permits one hour for
group meetings.
1.9.4 Cisco WebEx

It is a video conferencing and interactive application. It consists of WebEx teams,
WebEx meetings, and WebEx devices. Through Cisco, WebEx screens could be
controlled remotely. It integrates with applications such as cisco and Microsoft teams.
[7]
1.9.5 WebEx Teams

It offers video conferencing, texting, and file sharing. It has an ongoing virtual
meeting space for teams that are working remotely and in-office. It has a provision for
a whiteboard. It supports hybrid and on-premises systems even though it is primarily a
cloud-based service.
1.9.6 WebEx Meetings

WebEx meetings support large-scale events like webinars and virtual training. Around
3,000 participants can join in interactive webinars and events supported by WebEx
events. Large-scale virtual gatherings with more than 40,000 attendees can be
supported via WebEx Webcasting. WebEx meetings extend a customer support
service named WebEx Support. Live chat, video conferencing, remote desktop
18
management, intelligent queuing, and routing are some of the features given by the
service.
WebEx Devices
A kit of devices with a camera, codec, speakers, and microphone for small and
midsize meeting rooms is available.
Through Cisco, WebEx screens could be controlled remotely. It integrates with
applications such as cisco and Microsoft teams.
1.9.7 Go To Meeting
This is one of the tools that is widely used for online meetings, teleconferencing, and
sales demonstrations. This application is most useful to organizations and academic
institutions that have employees or students all over the world. Seminars and
presentations are effectively handled with this tool. Toll-free numbers could be given
to individuals who wish to set up a phone call.
Thus, the attendees will not incur long-distance charges. This is the unique feature of
the Go To Meeting. The drawing tool is one more important feature that helps the
presenter draw and highlights text on the screen while delivering the presentation. [8]
1.9.8 Go To Webinar
It is an extension of the Go To Meeting that has a registration feature. In addition, it
supports in-session tools like surveys and polls, which are best for generating ideas
and helpful for reaching across long distances.
1.9.9 Go To Training
It gives the user the ability to design and display a comprehensive list of courses for
students along with the cost of the course. Apart from this course material are also
stored for students so that they can download them.
1.10 Challenges in continuing Education:

In March 2020, the abrupt closure of schools, colleges and universities resulted in
termination of offline classes as well as exams and students were forcefully sent
home. Educational Institutes across the world were empty and administrators were
struggling hard to prepare for the “what comes next” situation.
19
Coronavirus pushed educational sector to make tough choices. Though most of the
institutes managed to shift to online education system but still online teaching
learning is no substitute to the “bricks and mortar” on campus learning experience.
In India, students were sent home in mid of March when its exam time for almost all
the final year students of all the courses. Thus, the conduction of examination,
assessment and evaluation were the immediate concerns. Even though alternate
systems were formed to conduct examinations fundamental questions were raised
like, is the degree valuable? Even if students get the degree to what extent, it will be
helpful for them in long run and for placements.
Especially for the professional courses like engineering and medicine where the
practical knowledge is important such questions were obvious. For the first year and
second year students classes were moved to online mode. From Kindergarten (K.G) to
Post teachers, students and parents faced graduation (P.G.) different challenges. Many
schools colleges postponed or cancelled the examination during the first wave of the
pandemic.
1.10.1 Financial Challenges for Institutes:

The financial planning of the education system was disrupted. Most of the experts
predicted three possibilities. First, one could be that the pandemic would be shorter-
In that case, there was no guarantee that all the students and teachers would return to
offline classes. Thus, this would had led to smaller classes and less revenue
generation.
The second possibility was if the authorities would face the pandemic continued in
upcoming semesters better arrangements were required to be made for online classes’
in that case more severe financial situation as more students would demand financial
relaxations.
The third possibility predicted was a toggle between online and offline classes and
this was the one that we all experienced. India experienced the first wave in March,
which continued until September–October 2020. The second wave hit around end of
February 2021 and showed devastating situations in April 2021. Around January
2022, the third way hit the country. Many medium and lower ranked institutes
suffered most. It was also predicted that some public state universities of higher
education could also be at financial risks.
20
1.10.2 Access to online Education:
Fig. 1.2 Online Education Access [13]
1.10.3 Digital Divide

Around 25 to 30 years back we were introduced to Internet. Internet access,
availability of electricity, technical skills these were the issues that we were also
facing prior to COVID. Digital divide was a concern before COVID and now during
the pandemic the problem has been aggravated. According to expert observations
there could be categories specifying connectivity and access.
1. There were students who were not at all connected.
2. There was a range of students who were not meaningfully connected. As they
could not perform daily transactions and shopping online, could not avail medical
help, educational facilities to the fullest because of many reasons like slow
internet speed, frequent disconnections, poor audio video etc.
21
3. Then we could observe people around who could connect but could not handle
excess information. They lost the motive as to why they were connected.
4. There were people who were not digitally literate. Thus, they could not turn
themselves productive. They became passive observers. Thus, they were
connected but functionally unconnected.
5. Lastly, out of all these we could find some who were meaningfully connected and
could achieve the objectivity of connection.
Youngsters often know more about technology that results into a digital divide
between them and senior citizens. Apart from this geographical location is also one
of the issue. The major issue is the cost to access. The high cost to access and connect
to the internet has kept most of the population away from this digital reformation.
Accomplishing internet connectivity through broadband is the most prominent
infrastructure challenge that we are facing today.
Infrastructure alone is not going to resolve the issue; supporting initiatives should be
taken for those who are already connected through broadband networks. Participative
initiatives could be helpful. Donation of used and functional devices and monitoring
whether they are reaching to the people in need of those devices. We cannot say that
bridging the gap to overcome this digital divide is the sole responsibility of any
government or some private investor or some social organization.
1.10.4 Effect of digital divide on mental and emotional health of students:

Mental and emotional wellbeing always has been a key player in the academic
achievement of students. Home and school environment, Family background, literate
parents and many other factors directly or indirectly affect the skills of the child. In
times of COVID, the mental and emotional health parameter emerged as something,
which was to be supervised frequently.
It is high time to decide as to what level of control we are going to give to technology,
connectivity, and information access. If we fail to decide upon a proper mechanism to
control unnecessary technology usage then within no time digitization as a necessity
in no time may overpower human beings. Due to lack of proper digital devices and
access, students started losing interest in subjects. It was very difficult for them to
acquire practical skills. Experiential and collaborative learning are very important for
the children.
22
Peer learning, sharing problems with friends and many small things students missed
which the indirect factors were making them mentally disturbed and lonely. Adding
to this the fear of getting infected, or handling some family member who was infected
were also some of the reasons for poor mental health. The fear of lack of medical or
financial help. There were students who could reach their hometown especially in the
first phase of lockdown. It was equally challenging for students as well as parents of
such children to face this situation.
1.10.5 Curriculum curtailment

Curriculum curtailment was a temporary measure but the prolonged closures of
institutes extended online education. In the initial stage students adjusted to the online
mode but later on with online consequent semesters, they started losing interest. This
was due to a lack of face-to-face learning experience and a lack of socialization.
Some universities decided to reduce the syllabus and curriculum so that important
topics could be discussed in detail and time could be spent on conceptual
understanding. Shortening the curriculum was not a full-fledged solution; it was just a
step to bring about some relaxation. Later on, the hybrid model was practiced in many
places. Soon, it was realized that this would be a new normal and will require
curriculum restructuring. In the era of post COVID, though we shifted again to offline
education it is no more the same situation.
There is a need to design a curriculum that could create a conducive environment and
will help students in acquiring technical skills and theoretical knowledge equally well.
1.11 Parent’s Role during Pandemic:

We all know that parent’s involvement in their child helps children to achieve better.
We are also familiar with the triadic nature of education system where the
collaboration between teachers, students and parents can yield magnificent results.
COVID-19 put forth a major transition into parent’s role. Earlier before the pandemic
and even now parents are concern about the learning outcomes of their ward and so
are keen to put them in best schools, hobby classes and would always ensure that they
would give them excellent facilities of whatever possible.
The pandemic period turned out to be a revelation for parents. It made them
rediscover their duties towards children. It forced challenging responsibilities on the
parent. The major one was to take care of the mental and emotional well-being of the
23
child. It was necessary to understand that the reactions to the adverse situation may
vary with children of different age groups. This could be handled by an emphatic
approach towards children. It was also important to set limits when needed.
There was a need of responsible caretaker in these times. Especially teenagers require
parents to spend quality time with them. Helping children in taking certain decision,
keeping them engage in doing some constructive work, playing with children and
giving them leading roles, were some of the factors, which helped children to release
stress. It was the responsibility of parents that isolation should not lead to social
disconnect.
Writing letters, emails, video calls, chatting with friends and relatives was equally
important to maintain the balance. Engaging students in physical activity to keep them
fit was also required. In spite of all this, there were children who found it difficult to
accustom to the new normal. Anxiety, depression, lack of self-care, irregular eating,
and sleeping habits were the negative effects that some children were facing.
The habituation to these habits was not good and needed consultation of experts. The
loss of friend, or relative or fear of such situation were the major causes of anxiety
and distress. The uncertainty that prevailed about education and their career path
added to the mental imbalance and mood swings especially in youth.
1.12 Teacher’s role and challenges:

Teachers were supposed to play multiple roles before COVID. The scenario of
COVID demanded more multifaceted role from the teaching fraternity. The abrupt
transfer to remote learning left teachers with no choice but to adapt the virtual system.
Teachers were not trained for this and had to rethink the lesson plans and structure for
the online lectures. They had to figure out how to design presentations, conduct
practical laboratory sessions, and teach art and craft in virtual mode.
Most importantly, teachers could not get student feedback and thus were unable to
interpret whether learning was taking place. Most of the time students were reluctant
to put on their videos; while sometimes the teachers and students were not reachable
due to technical limitations, some were hesitant to ask questions, like this, there were
many issues related to online teaching. For practical subjects, it took more time for
students to understand the concepts.
Unlike the traditional method, teachers could not move around in the classroom and
check with each student if they understand the subject. The shift took place at the end
24
of the academic year when in most countries it was the examination period. Thus, the
teachers knew the students and vice-versa. Hence, it was easy to communicate and
even it was possible to evaluate students on the overall annual performance. Final
grades were given and students were promoted to higher standards.
The major concern was the conduction of entrance exams for admission in premium
institutes for professional courses. These entrance exams, which take place at the
national level, were postponed. Again, the role of teachers was extended, as they had
to revisit and had to chalk out the exam procedure. In addition, they had to reset the
question papers. Following this, the start of the next semester and in most places the
beginning of the new academic year was troublesome. It was equally challenging for
teachers and students.
The first-year students of all courses were not familiar with the rules, regulations, and
teachers of the institutes. They were introduced to colleges or schools in online mode.
Thus, starting the session remotely was the biggest challenge for the faculty. After
this, there came a period when teachers had to toggle between the online and offline
modes of teaching. This added to the uncertainty and teaching preparations.
After the second wave, decisions were taken by the local and central governments to
reopen educational institutes. However, due to the unpredictability that prevailed
students were not willing to attend offline lectures. Parents of outstation students were
not ready to send their ward for physical classes as hostel accommodation, health
related insecurities still existed. As a result, teachers had to shift to a hybrid model
(both online and offline options were kept open). It just added to the overhead for the
teachers. [6]
1.13 Challenges faced by students
1.13.1 Screen time

These days we find everyone lost in digital space most of the time. It could be online
meetings, entertainment or simply browsing the social media for long hours. There is
no scrutiny as to what messages and content anyone is sharing on the social networks
nor there any restrictions on to what to watch and for how long. It is important to
notice how all this is affecting the human brain.
Study conducted by neurologists explained that continuous browsing activities
activates reward centers found in the mid brain. Further, the pleasure-based centers in
25
the cervical part are turned on. These centers then encourage the individual to watch
the next video or may be the next episode of the series.
The frontal part of the brain is quite developed in human beings and initiates rational
thinking. One could get signals from the frontal part of the brain that one should stop
watching. This creates constant struggle between the front and the mid brain leading
people to unnecessary exhaustion. [15]
1.13.2 Reading Skills

Reading text, stories, news, magazines is a very basic ability which education
promises to imbibe among children. In addition, a basic element takes anyone towards
literacy. The 2022 ASER report notified about a major decline in the reading ability
of the students. The grade three students of academic year 2022-2023 were found to
be the affected at large. These students were in Kindergarten when the pandemic hit.
They missed on the foundation years of their schooling.
In the LKG, UKG and first grade students follow the instructions given by teachers.
Rather they rely on classroom instructions and follow it sincerely. Thus, it is to make
them understand the foundational skills. The decoding of the words and the phoenix
awareness was the part, which was missed by these students in the pandemic years.
The result is though they are in 3rd grade they cannot read the book of grade 2nd.
In the hybrid mode when they sometimes took offline classes, they were still
following social distancing and both the teachers were wearing masks. This again
restricted the understanding. [9]
1.13.3 Writing Skills

In the pandemic phase this was the skillset in which the lacked largely. Children of all
age groups faced different problems with writing skills. Firstly, there were the pre-
primary student in the age group of 6-10 years who had not acquired the reading
skills. The basics like how a child should hold a pencil or how to put the finger
pressure on it to be able to write an alphabet could all by the teacher in a classroom
and that too with an individual attention. Thus, these students lost their major years
when they were supposed to be trained in these skills.
Next, there were students of middle school, high school and even college students
who lost the habit of writing. In college students, the habit of taking notes during
lecture was completely lost. Instead, they preferred to take photos of the presentation
26
slide. In addition, they were ready to submit hard copy of the word document or a
PPT as it was easy for them to search from google update it and take a print of it.
Thus, writing in their own words and doing submissions was lost habit for them. [9]
1.13.4 Knowledge Retention

There is a lot of difference between hearing and listening. If one listens to something
then there is a chance of information gain. Then eventually one should think on that
information or process it to understand the concept. Once the matter is understood
then only it will be retained. The number of times one applies it chances are there that
the concept is retrieved repeatedly and hence gets permanently stored in the memory.
In lockdown, students faced issues while attending lectures in online mode. Some of
the students were not comfortable sitting at one place for long. Some felt anxious, as
they had to keep on their videos. Owing to this, these students sometimes moved
around, or were not ready to put on their camera. Many of them could not catch with
the long screen time. Some were having disturbances at home. While some were
agitated and jiggled. The students reacted as they were facing challenges in focusing.
The teachers interpreted this as lack of student engagement.
Thus, there was a need to engage students with a different approach. Like asking
students to share their work for example how they were working on a project or in
case of schoolchildren how they were able to read or understand a concept. Being in
touch with each of the student individually was one of the major attribute that help
students in focusing on online classes. [14]
1.13.5 Offline Mode

 In an offline face-to-face conversation, we can observe the body language of the
person, we are interacting with. Some important elements like eye contact and
smell were missed in online communication. These factors also contribute to
understanding the personality of the individual. Children in the pre-primary
section learn a lot through activities. The participative teaching-learning method is
suitable for these kids. Storytelling, enacting, and similar types of engagement
help them learn better. Through this, they understand the way to connect with
society. Also, conversations with teachers and parents help nurture their
behaviour. Right from the birth of the child until the age of around three years, the
child develops vocabulary and learns languages.
27
 According to Vygotsky theory, language is the basis of learning. It accelerates
reading, writing, and reflective thinking skills. The creation of a thought process is
deeply rooted with the belief system of an individual. It further shapes the
behaviour and attitude of any individual. The belief system one develops by
looking at people around or the experiences that life offers to a person. These all
aspects are developed in a child from pre-primary to primary level. This could not
be achieved in an online setup effectively.
 During online instructions, students developed the habit of just logging in for the
lecture and then moving around doing multiple other things or household chores.
The disciplined environment and the mind-set needed for sitting in a classroom
and getting knowledge faded gradually during this pandemic period. Reading and
writing skills were also affected. Especially by the time a child completes the
primary section, he / she is expected to acquire basic literary and numeracy skills.
 After a gap of two years when the students started resuming offline classes it was
noticed that they were struggling with lack of confidence in oral exams, seminar
presentations and similar such skills. Likewise, there is a setback in the writing
habit of students. They preferred to click photos of the study material shared by
the teachers during class or take screenshots instead of noting down manually.
Worst was the condition of primary or pre-primary section who had lost the major
years of their education when they were supposed to learn the basic reading-
writing skills. Many Government and private organization to identify the learning
loss and came up with initiatives to improve the situation carry out surveys.
 The National Achievement Survey (NAS-2021) and the Annual Status of
Education Report (ASER) are the two major surveys that explained well the after-
effects of school shutdowns in rural as well as urban parts of the country.
According to NAS, Delhi was one of the five states where standard third students
reflected lower performance. Whereas, 8th-standard students did very well. In
addition, it was found that the average result dropped down by 7% from the last
NAS survey in 2017 to the latest one in November 2021. [9]
28
Fig. 1.3 Challenges of Students for Online Learning [13]
 Through the ASER survey, it was observed that only 27% of students studying in
the third standard could read previous years' textbooks, and that too with a lot of
effort. The survey came up with the fact that around 48% of fifth-standard
students were able to read books in second grade.
 The Karnataka government came up with a program named “Odu Karnataka” to
promote better reading skills in students. The state found that the reading habit of
the students was affected during the pandemic and hence proposed this program.
29
Fig. 1.4 Online Access in Rural Areas [13]
1.14 Problem at hand

The COVID19 pandemic significantly exposed the students to the digital world. Until
then they were largely learning in an offline and classroom mode. Thus were having
much higher level of self-discipline while operating the mobile devices and had
higher level of control about when, where and to what extent leverage the mobile or
the technology. However, during the pandemic times and especially in the lockdown,
it was mandatory to use the digital devices for almost all the activities and daily
chores. This over exposure to digital devices made the children to develop a mindset
that everything can be performed at the click of the mouse/ button. This caused a lazy
attitude within many of them.
On the other hand, there were children who took this in a positive stride and utilized
the lockdown situation to upgrade the skillsets in the areas of their interest.
The objective was to understand to what extent and in what way the children have
been affected by the pandemic. Another goal was to find what difference COVID has
brought in the education system, approach of the parents and teachers, policy makers
and academic stakeholders.
30
In the upcoming years what should be the comfortable mode of education. Most
importantly, it was also necessary to understand the perspective of the students
towards this new way of learning.
One more angle to this research is to know the extent to which the pandemic has
affected their physical, mental, temperament, co-curricular and extracurricular
activities.
It has been observed through previous research work that the personality traits
contribute significantly in deciding the academic progress of the students. The
pandemic affected the mindset of the students, their growth in several aspects.
The above-described scenario created an opportunity to do research in this area. The
researcher has tried to reveal the effect of the pandemic on the various dimensions of
the students’ personality. In addition, the work could contribute in gaining insights
from students. These inputs could help in framing a robust education system which
can withstand any pandemic or similar other situation which the upcoming generation
could implement if they come across similar adverse situation in future.
1.15 Objectives of Research

1. To study the overall impact of a COVID-19 pandemic on students of different age
groups.
2. To study attitude, temperament, motivation, and learning styles.
3. To analyze deep learning-based algorithms for predictive analysis of sentiments
and behavior of students.
4. To analyze and identify face recognition algorithms to capture the real-time image
for the prediction of the behavior of students.
5. To study the impact on various parameters in academics, co-curricular and
extracurricular activities.
1.16 Hypothesis
H1: There is an impact on the attitude, temperament, motivation, and learning styles
of students due to the COVID-19 pandemic.
H0: There is no impact on the attitude, temperament, motivation, and learning styles
of students due to the COVID-19 pandemic.
H2: There is an impact on academic, co-curricular, and extracurricular activities due
to the COVID-19 pandemic.
31
H0: There is no impact on academic, co-curricular, and extracurricular activities due
to the COVID-19 pandemic.
1.17 Need of Study

The COVID-19 pandemic disrupted the educational system thereby hampering the
physical, mental, and behavioral growth of the students at all levels. There were many
concerns raised at the end by parents, teachers regarding the academic and overall
well-being of the students. Teachers tried to pave way for better evaluation systems.
Students and everyone related to the educational field took time to adapt to the sudden
change in the educational system.
During the first lockdown period, it was observed that students took it on a lighter
note and considered it a temporary change. Though COVID 19 was declared a
pandemic, it took time for everyone to accept and understand the serious replications
of the same. Gradually when the pandemic prolonged creating adverse situations
around
On one hand, the teachers and students were being accustomed to the new way of
teaching learning with various online platforms like zoom, Microsoft teams, and
Google-meet. On the other hand the fear of being infected by the virus, and the
pressure of following the norms imposed by COVID-19 resulted in stress.
1. Due to the COVID pandemic, students were at home for around 1.5 to 2 years.
2. The pandemic imposed social isolation on students so there is a need to study
their mindset.
3. There was a complete shift in the educational paradigm so need to understand the
implications in various areas.
4. To come up with common issues and solutions to make it a new normal.
1.18 Scope of Work

There is a broad scope to study the subject because so many factors were raised
during the COVID pandemic in the field of education.
1. Unavailability of infrastructure at different locations for online mode.
2. Unavailability of skilled people in schools and colleges for using online teaching,
learning, and evaluation systems.
3. Unavailability of resources such as computers, laptops, mobile phones, and
internet services.
32
4. Loss of jobs by parents.
5. The proposed model will be implemented in schools within PCMC, and colleges
within the periphery of Savitribai Phule Pune University.
6. The geographical area is one of the restrictions implied.
7. Thus, there are so many issues mentioned above but we have restricted our study
to identify the motivational, behavioral and physical impact on students due to
COVID pandemic.
As per the existing schools and Colleges under study
Universe
01 Schools (PCMC) 340
02 Jr. Colleges and Sr. Colleges under SPPU 990
Table 1.1: Universe and Sample for the proposed Study

Amongst available institutes, the researcher is planning to undertake the following
sample for the study
01 Schools (PCMC) 50
02 Jr. Colleges and Sr. Colleges under SPPU 100
Table 1.2: Sample
1.19 Thesis Outline

The thesis comprises of seven chapters. The overview of each of the chapter is given
below:
Chapter 1: Introduction
The chapter introduces the topic of the research study. It explains the background of
the study. The evolution of education and then the sudden change brought about by
the pandemic in the academic world has been explained. The place of ICT tools in
education is also discussed. The major Schemes and resources announced by the
government under educational provisions were detailed. The challenges faced by the
students, teachers, parents are elaborated. The approach of students’ towards studies,
the effect on gaining the basic skills is also mentioned. The objective of the research,
33
Hypothesis of the study, need of the study, and scope of the work has been stated.
The thesis outline is given in this chapter.
Chapter 2: Literature Review

In this chapter, the work done by the researchers related to the topic under study was
reviewed. The research papers that contributed in the machine-learning and deep
learning algorithms were considered. The survey papers were also studied. Various
frameworks, application based papers were reviewed. The literature related to
students’ psychology, impact of personality factors on academic growth were studied.
The influence of previous pandemic on educational field and the strategies used to
overcome the loss were assessed through the research papers. The previous research
work contributing in the role of technology, recent computer science methodologies
and the data science tools were explored to gain Knowledge about them.
Chapter 3: Machine-Learning algorithms and technology for model formation

The chapter briefs the concept of Machine-learning and deep learning. The role of
AI, its evolution has been explained. The various applications of AI and its
significance in transforming education is discussed. The machine-learning algorithms
used in the research work are detailed. The implementation process flow of all the
four ML algorithms are explained. The terminologies of ML and AI approaches and
their conceptual explanation is also given in the chapter. Major classification of the
machine-learning algorithms is also explained.
Chapter 4: Facial expression recognition model

Emotion recognition through facial expressions is the second module of the research
Expressed in this chapter. The working of Haar cascade algorithm is also detailed.
The steps of video capture and image processing is also explained.
Chapter 5: The student survey application tool

This chapter focuses on the application tool, which was developed to analyze the
physical, behavioral, motivational, co-curricular and extracurricular features of the
students. The working of the proposed model, the conduction of the survey both have
been explained. The implementation of the application, the user and the admin view
are given. The framework for the application development, its components, and
34
advantages are mentioned. The programing language, libraries, the architecture of the
framework are discussed.
Chapter 6: Results and Discussion

The evaluation of the proposed model is done in this chapter through the results and
discussion. The results are analyzed to understand the impact of the pandemic on the
students of different age groups belonging to various schools and colleges. The results
are presented in graphical format that helps to visualize it easily. The actual responses
of the students collected by the survey application are examined. The comparative and
individual analysis of the machine-learning models is projected with the help of
confusion matrix.
Chapter 7: Summary
The crux of the complete research work is reflected in this chapter. It summarizes all
the phases of the research model. It evaluates all the modules. The observations are
put forward. The conclusions drawn from the study are stated. Based on the
conclusions the limitations of the research study and future scope of the project are
compiled.
1.20 Summary
This chapter gives a brief introduction, background and need of the study. The
COVID-19 pandemic brought a transformation in the academic world. In relevance to
this, it was interesting to explore the impact of the pandemic on the students’
personality. There are many attributes to a students’ personality. The objectives of
the study, hypothesis, and scope of the study are described in this chapter.
35
CHAPTER 2
LITERATURE REVIEW
2.1 Introduction
The chapter briefs us about the extensive work that has been carried out until now in
the area that has been chosen for research. A review of the literature helps the
researcher to summarize the existing knowledge. It lays a foundation for the current
research work. It helps to identify and address the potential gaps in the previous
research. It uncovers the thrust areas of the research topic. In addition, it avoids
duplication of information. It reveals the theories, and contributions made by
researchers.
The topic undertaken by the researcher demanded a literature review from different
domains like education during COVID-19, machine-learning, and changes in the
physical, mental, and behavior of the students due to COVID. To build the proposed
model research papers related to the application of various machine–learning
algorithms were studied by the researcher.
Literature related to student psychology, behavior, mood swings, physical health,
socialization, and motivation especially during COVID pandemic was studied. The
effects of the previous pandemic on education were studied. The sudden growth in the
use of ICT tools and the various platforms that gained popularity was reviewed.
The second part of the model demands a literature review to be carried out for
different deep-learning algorithms used in facial expression analysis. How the
computations are to be made to categorize different emotions was analyzed. The
Django framework applications were understood through literature.
An application has been framed in the research work as an interface between the user
and the model. Thus, literature related to it was considered. The literature review is an
essential step in the process of research. It helps us to acquire relevant information
and we get a direction for carrying out further research. It identifies the need for
additional research in the field of the selected research subject.
Various research papers, journal publications, and books were reviewed for the
research work.
2.2 Literature Review
36
Nicu Sebe et al. (2004), in “Authentic facial expression analysis”, developed a
database for capturing the real-time facial expression of the participants. The system
was created to uncover the real state of mind through expression recognition.
Participants were made to view a video. The video contained clips from recent
movies. Horror, Comedy, and a variety of movie clips were used to capture the
emotions of the participants.
From around 60 participants about 28 students from computer science willingly
contributed to the database. After the emotions were analyzed, the second step was to
interview these participants to gain an insight into their real state of mind. The
emotional analysis was done in an unobtrusive method using a hidden camera. Many
Machine-learning algorithms and classifiers were applied to classify the facial
expressions into appropriate categories.
Generative Bayesian Networks, SVM, KNN, PEBLs, and CNN were used. An
adaptive boosting, the voting algorithm was used to improvise the accuracy of the
classifiers. The researcher was successful in generating an authentic facial expression
database. The database was designed to make it available to the scientific community.
The classifiers for expression recognition were validated using this authenticate
database as well as the Cohn-Kanade database. Finally, the face tracking system and
the classifiers were merged to develop the real-time facial expression recognition
system. [18]
Thomas A.Litzinger et. al. (2007), in “A Psychometric Study of the Index of

Learning Styles”, introduced a modified Felder-Soloman Index of Learning Styles
(ILS) by changing the two–choice scale to a five-option response scale and adding
moderate preference. The four learning styles Active-Reflective, Sensing-intuitive,
visual-verbal, and sequential global were used.
The basic ILS was modified to find out whether it improves reliability and validity,
whereas the basic ILS was also used to examine the reliability, factor structure, and
construct validity. The author compared the basic ILS and modified ILS and
concluded that the reliability and construct validity from both factor analysis and
student feedback for basic ILS were significantly good.
However, the modified ILS with the five-option response scale reflected a significant
improvement in the reliability for three out of four learning styles but did not
introduce any notable change in the factor structure and student perspective validity.
37
Considering the results of the study the author has decided not to move ahead with the
modified ILS. [19]
Adel Ben Youssef et. al.(2008), in “The impact of ICT on student Performance in
Higher Education: Direct effects, Indirect effects, and Organizational Change”,
studied the influence of ICT on student’s academic performance by considering
direct, and indirect effects of ICT and the extent of organizational change needed for
achieving better performance.
In the first half of the paper, the author has reviewed the literature that explains the
direct effect of ICT on higher education and student performance. In the second
section of the paper, the author focuses on traditional key factors such as student-
teacher characteristics and educational environment and how ICT has influenced these
factors, which may affect student performance. The third section reflects on the lack
of organizational change and its effect on student performance.
The author concludes that the use of ICT in education has a magnificent impact on
students' performance as they are acquiring collaborative skills and thus ICT –enabled
education is helping to bridge the gap between Industry and academia. The author
further imposes the need for organizational innovations and the adoption of best
practices and policies by an organization to gain better academic performance by
students. [20]
Mehdi Sagheb Tehrani (2009), in “The Results of Online Teaching: A Case Study”,
conducted a case study to explore the advantages and disadvantages of online
learning. This study was carried out on some students in the MBA program. The study
revealed that online education has many advantages the prominent one being the
flexibility that it offers. The result of the Case study reflected that there is tremendous
growth in student enrollment for higher education online courses over the past few
years.
The author concluded that online learning is very well accepted and it is no more a
new concept. Online learning provides a strong platform for academicians and
learners to exchange ideas, and concepts and work together 24*7 from anywhere
around the Globe.
The author emphasized the need to strengthen certain important attributes of online
teaching-learning platforms such as faculty recruitment and training, organizational
38
structure, student service units, and promising employees for maintaining
infrastructure. The author promotes online learning but questions the percentage of
practical-based courses to be added at the Under Graduate or Graduate level. [21]
M.P. Walker (2009), “The Role of Sleep in Cognition and Emotion”, studied how
sleep was associated with cognitive skill development. Several sleep-related models
were reviewed and it was observed that long hours of sleep do not strengthen memory
but help in summarizing knowledge from generalized emotional experiences.
The researcher comes out with a theoretical framework to analyze the functioning of
the brain related to sleep hours. The model aims to find how sleep affects the mood
and emotional patterns of the human brain. [22]
Sucheta V Kolekar et. al. (2010), in “Learning Style Recognition using Artificial
Neural Network for Adaptive User Interface in E-learning”, attempted to recognize
the learning styles of students by monitoring their actions of students during their
usage of E-learning applications. The author has designed an E-learning Model by
referring to the Felder and Silverman learning style model for classifying students.
The learning style recognition is based on Machine-learning algorithms prominently
Artificial Neural Networks and Web-usage mining.
The learning methodology proposed in this paper helps to identify various learning
patterns, which are further used to develop an adaptive user interface that tries to
intensify learning efficiency by customizing the learning contents according to learner
preferences. [23]
MikkoVesisenaho (2010), in “ICT Education and Computer Science Education for

Development-impact and Contextualization”, has studied and summarized the
research on contextual approaches in Computer Science and ICT education in Africa.
According to the author, contextualization emphasizes context-sensitive planning and
provides a flexible approach, which is very much needed by students and teachers for
facing the challenges in delivering ICT-enabled education. In addition, Context-
Sensitive planning provides global scope and ensures long-term sustainability in the
field of computer science and ICT education. [24]
Renu Nagpal et al. (2010), in “Hybrid Technique for Human face emotion
Detection”, concentrated on classifying four emotions of anger, happiness, fear, and
39
neutrality. A database from noisy salt and pepper environment was used for training
the dataset. The combination of Adaptive median filter and mutation BFO helped in
removing noise from the face images.
Back propagation neural network and radial basis neural network were implied for
deciding the category of the identified emotion. Less number of epochs were required
for radial basis neural networks as compared to other neural networks. Thus, it was
noticed that the proposed method was suitable with high noise of around 90%.
Comparative analysis showed that the proposed method was able to recognize
expressions and emotions successfully even in a noisier environment. [25]
Jagdish Lal Raheja et al. (2010), in “Human Facial expression detection from
detected in the captured image using back propagation neural network”, introduced a
straightforward architecture for recognizing facial expressions. The method is simple
to adapt in a real-time system. A brief description of the methods for web cameras
such as picture capture, face detection, and image processing was provided. Viola and
Jones used Add Boost Haar classifier technique were used for face and emotion
detection. [26]
PlamenZahariev et al. (2013), in “ICT Convergence Challenges in Education and

their Impact on both Instructors and Students”, have briefed about the emergence of
ICT. The advantages and challenges faced both by teachers and students with the
advent of ICT are studied in this paper. The merging of ICT with education has
established a need for the educational sector to upgrade technological platforms. The
author has summarized certain modern technologies such as cloud technology, pipe
technology, and device technology.
According to the author, the merger of ICT in education puts forward many
challenges to ICT, but it will help both the industry and education sector to evolve.
The author further assures that the upcoming technologies and their integration with
ICT will lead to the betterment of the Educational Sector. [27]
Wenbin Liang et al. (2013), “Sleep duration and its Links to Psychological Distress,
Health Status, Physical activity, and Body Mass Index among a large Representative
General Population Sample”, interpreted the relationship between various physical
and mental health parameters. It was reported that people who were actively involved
in physical exercises and were free from mental stress were satisfied with less than 6
40
hours of sleep. In addition, individuals with a BMI of 25 or more required less sleep.
The data from National Health Interview Survey was utilized for the analysis.
Multinomial logistic regression was used in the research. [28]
Shuiwang Ji et al. (2013), in “3D Convolutional Neural Networks for Human Action
Recognition”, had taken into account the automatic detection of human activity in
surveillance footage. The majority of the existing techniques train their classifiers on
sophisticatedly designed features calculated from the raw inputs. Convolutional
Neural Network (CNN), which belonged to a specific class of deep learning model, is
capable of acting directly on the input data.
It was observed by the researcher that models currently could handle 2D inputs. In
this research work, a novel 3D CNN architecture has been framed for action
recognition. By using 3D convolutions, the model pulls features from both the spatial
and the temporal dimensions. For this, it captures the motion information stored in
numerous neighboring frames.
From the input frames, the generated model creates numerous channels of
information, and the final feature representation incorporated data from all channels.
After implementing the model in the real-world environment of airport surveillance
videos it was found that this new model performed well and represented good results
as compared to the baseline methods. This proposed model was evaluated on TREC
video retrieval evaluation that consisted of surveillance video data recorded at London
Gatwick airport. [29]
K. H. Wanjale , et al. (2013), in “Use of Haar Cascade Classifier for Face Tracking
System in real Time Video”, discussed the face identification and tracking system
with real-time video in this study. The system was split into two major components.
To identify a face in the picture Face Name Graph matching algorithm was applied.
The algorithm consisted of numerous techniques like the Haar-cascade approach, and
the OpenCV library.
The clustering technique was utilized to follow the video’s face. The system could be
implemented for security purposes. The video recordings and questionable activities
were fed into the system as input sources. From the input video streaming, one or two
individuals were identified and tracked. To detect the face in the video Haar-cascaded
41
classifier was used. Face-clustering algorithms were used for tracking. The database
connectivity was established by linking .net framework with MS-access. [30]
Geraldine Gray et. al. (2014), in “A Review of Psychometric Data Analysis and
Applications in Modelling of Academic Achievement in Tertiary Education”, has
reviewed and focused on relevant psychometric factors such as Cognitive ability,
personality, motivation, and learning strategies and their influence on the academic
performance of tertiary students.
The objective is to identify students at risk of failing or dropping out of the course so
that early attention could be given to such students. The big five factors under
personality evaluation such as extroversion, neuroticism, openness, agreeableness,
and conscientiousness are considered and significantly high correlations are found
between openness and academic performance but very low correlations were found
between agreeableness and academic performance rather agreeableness is correlated
with absenteeism in the first-year students.
There are two major types of motivation- outcome expectation and Self-efficacy. This
study reflects that self-efficacy and intrinsic motivation have more influence on the
academic performance of tertiary students. Learning strategies like deep learning,
shallow learning, self-regulated learning, and strategic learning were evaluated out of
which it was found that self-regulated learning has a high correlation with academic
performance.
The author has compared the results of various data mining algorithms like SVM,
Decision tree, and Bayesian network for predicting the academic performance of
students in tertiary education. The author concludes that there is a need to come up
with better data modeling techniques so that a customized personality-based predictor
of academic performance of tertiary students can be achieved. [31]
Charles Dziubanet. al. (2015), in “Student Satisfaction with Online Learning: Is it a

Psychological Contract?” attempted to discover the probable relationship between the
theory of psychological contracts and student satisfaction with online learning.
Engaged learning, agency, and assessment are the three major factors that are
considered for measuring student satisfaction with online learning.
The Research initiative for teaching effectiveness uses a five-point Likert scale
ranging from strongly agree to strongly disagree. A longitudinal study has been
42
carried out and Google form was used to collect the opinion of students about online
learning. The factor correlation matrix was evaluated. The overall satisfaction resulted
in .74 by the Kaiser method. Pair-wise combinations were represented that proved
significant in the Bonferroni comparison model.
The author concludes on the note that studying the psychological attributes of students
helps to identify the state of students before commencement of the course and student
satisfaction is something that we try to find after the course completion hence by
correlating these factors efficient learning models can be designed. [32]
Simin Ghavifekret. al. (2015), in “Teaching and Learning with technology:

Effectiveness of ICT integration in schools”, focuses on teachers’ points of view on
ICT integration in schools and its impact on the overall teaching-learning process.
The quantitative research methodology was used to collect responses from teachers
and the data was analyzed using Statistical Package for Social Sciences (SPSS). 101
teachers from ten public secondary schools in Malaysia were considered as the sample
population under study.
The research findings show that the ICT-enabled teaching-learning environment
improves the teaching-learning process thereby providing scope for effective
classroom management by teachers and imbibes better learning abilities in students.
The findings also reflected that areas such as effective teacher training, maintenance
of ICT-equipped classrooms, and updating technological services demand
improvement for implementing ICT in education. [33]
M Varun Kumar et. al. (2015), in “The role of ICT tools in teaching in India” puts
forward some major ICT tools and techniques such as mirroring techniques and
virtual classroom creation which ensure better teacher training.
The author suggests certain tools such as e-magazines and industry forums for
improvising sustainable learning and stresses the need for inculcating practices
amongst teachers such as creating lecture notes and posting them on specific URLs.
[34]
Shrutika C Dhargave et al. (2015), in “Survey paper on methods of emotion

detection from still images”, overviewed emotion recognition using a variety of
methodologies. The strategies for the same were briefed in this work, which would
43
assist in comprehending these methods quickly and effectively. One can select the
best-fit algorithm and utilize it for future assessment based on the person’s project.
Patch-based feature extraction, Matlab, and 3D model methodologies were applied for
feature extraction. Emotions identified from the extracted features were then
compared with the emotion database to determine the kind of emotion. This was
carried out with neural networks, deep belief networks, and pattern recognition
techniques. [35]
Havan Agrawal et al. (2015), in “Student Performance Prediction using Machine-

Learning”, studied the parameters of students' grades in secondary education, Living
location, and medium of teaching. This experiment was carried out to find if there
exists any correlation between these parameters and student performance.
A neural network algorithm was employed in the study. Three hundred students from
five-degree colleges were considered. For variable selection and prediction,
MATLAB was used. An application was implemented with a neural network
algorithm. The factors with a probability value greater than 0.70 were classified as
influential. The previous year's performance of the students proved to be the highest
influential factor. In addition, it was noted that the accuracy of the neural network
algorithm was directly proportional to the size of the dataset. [36]
Anna Qian Sunet. al. (2016), in “Online education and its effective practice: A
Research Review “, recommended certain best practices which will be helpful in the
efficient conduction of online teaching and learning. Through this paper, the author
has reviewed the evolution of online education and its impact on the emergence of
online course design. Further a few online education initiatives like Coursera,
eduMOOC, edX, Udemy are also discussed.
The author summarizes that in online learning attention should be given to the
coordination between cognitive abilities and teaching presence so that one can come
up with a suitable online approach for designing and delivering the course content.
The basic problem in online education is the creation of an interactive virtual
environment and their view poses the need to establish a good collaborative platform
so that the students will get a feel of the online community. [37]
Collen Halupa (2016), in “The impact of online learning and technology on student
Physical, Mental, Emotional, and Social health”, tries to inform/educate people about
44
the adverse effects of technology on teenagers as well as small children. The author
reveals that there is advancement in technology and it is high time that people need to
adapt to technology-based learning but also stresses the need to limit the exposure to
excessive use of technology.
It has been observed and reported by the author that extensive use of media and
technology at improper age and in a wrong manner causes hindrance in the physical,
mental, and emotional growth of children and can cause a long-term disparity in both
physical and mental health. [38]
Yaman Koseoglu (2016), in “To what extent can the big five and learning styles
predict academic achievement”, throws light on the relationship between learning
styles, personality, and academic achievement. The participants of the study were
undergraduate students from various streams like arts, science, engineering, and
architecture. Out of the Big, five factors of personality Conscientious and
agreeableness depict a positive association with all four learning styles whereas
neuroticism produces a negative impact on all four learning styles.
The research findings reflect a dynamic relationship between personality and learning
styles and represent a combined impact of personality and learning style on academic
performance. The study shows that learning style sometimes intervenes in the
association between personality attributes and academic progress. [39]
Fatima Zahra Salmam et al. (2016), in “Facial Expression Recognition using

Decision Trees”, used a geometric approach to extract features. The distances
between the eyebrows, eyes, nose, and mouth were calculated through Euclidian,
Manhattan, or Minkowski distance. These three distances were compared using the
Decision tree algorithm. The decision tree classifier distinguished the expressions into
seven types of emotions. The algorithm was applied to JAFEE and COHEN databases
with a precision of 89.20% and 90.61% respectively. [40]
Dr. Poonam Tanwar et al. (2016), in “Facial expression detection using hidden
Markov Model”, handled the issue of human face expression variation using the
Hidden Markov Model and K-Nearest neighbor. The KNN and Hidden Markov
Model are employed successively in the study. The comprehensive approach laid to
favorable results in evaluating the performance of the algorithms. The proposed
structure had two stages of Hidden Markov Model and was preceded by a layer of
45
KNN. The output was verified by testing the database containing 240 images for eight
facial expressions. Fifteen images were tested for the expressions of anger,
annoyance, sadness, surprise, happiness, and neutral. Seven images were tested for
grumpy and disgusting feelings, which were interpreted on the face. [41]
Ratna Astuti Nugrahaeni et al. (2016), in “ Comparative Analysis of Machine-

Learning KNN, SVM, and Random Forests Algorithm for Facial expression
classification”, compared and graded the effectiveness of three machine-learning
algorithms for emotion classification from facial attributes. Twenty-three variables
were derived from the feature distance of the face. These variables served as input
characteristics and resulted in seven categories of expressions. To train the model
certain test cases were framed. The amount of data for each test case varied. The
output of each algorithm with the smallest amount of data was quite pleasing with an
accuracy of 75.15 % achieved by KNN.80% by SVM and, a test case utilizing the
random forest method yielded a score of 76.97%. When maximum data was used, the
respective accuracy for KNN, SVM, and Random Forests algorithms was 98.85%,
90%, and 98.85%. Discriminant analysis was used to classify the training data for
each test case, with an accuracy rate of 97.7%. [42]
Andre Teixeira Lopes et al. (2017), in “Facial expression recognition with

Convolutional Neural Networks coping with few data and the training sample”, came
up with a sorted solution for facial expression identification. Certain steps for image
pre-processing were merged with convolutional neural networks. According to
experiences, CNN works well for big data. However, due to limitations in the
availability of public datasets, few features were extracted from the image. The
features were chosen based on their relevance in posing emotions. CK+, JAFFE, and
BU-3DFE databases were used in the experimental study. The methodology achieved
96.76% accuracy on the CK+ database. [43]
Hidayet Takciet. et al. (2017), in “Measurement of the appropriateness in career

selection of the high school students by using data mining algorithms: A case study”,
explores the association between academic achievement and students’ abilities,
interests, and expectations. Various predictive data mining algorithms like SVM,
CNN, Naïve Bayes, and MLP were compared to obtain a model that will help students
46
for selecting a better career option. The sample data was collected using
questionnaires and psychometric tests that were conducted on students.
The analysis of the experimental result showed that the questionnaire dataset proved
admirable in comparison to the psychometric dataset reason being the questionnaire
dataset considered many aspects of the students moreover the demographic aspect
turned out to be a major differentiating factor. It was found that the interrelation
between students ‘ability and success was more prominent rather than the relation
between students’ interest and success. The SVM algorithm reflected more accurate
results than the other three algorithms; also, the train-test method was more fruitful
than the 10-fold cross-validation method. [44]
Osisanwo F.Y et al.(2017), in “Supervised Machine-Learning Algorithms:

Classification and Comparison”, compared seven different supervised machine-
learning algorithms on several parameters like the speed of learning, speed of
classification, accuracy, tolerance to missing values, handling redundant values,
effort to deal with overfitting, attempts made for incremental learning, knowledge
transparency, dealing with discrete, binary or continuous variables. Weka machine-
learning tool was implemented on a diabetes dataset having 786 records with eight
independent variables and one dependent variable. According to the study, it requires
time to establish a model considering the accuracy, precision, and minimal errors. It
was concluded that an environment for distributed processing should be considered. It
is important to find factors that show a high amount of correlation between the factors
that will ultimately improve the model’s output efficiency. [45]
Muhammad Ahsan ulHaq et. al. (2017), in “Psychometric study of depression,

anxiety and stress among university students” uncovers the association between
depression, anxiety, stress, and the socio-demographic factors of students. Students of
various undergraduate courses were examined through the DASS-21 stress scale. It
was found that depression, anxiety, and stress were more prominent in male students
in comparison to female students. In addition, the study showed that residential status
and parents’ education were the key players in molding the attitude of students toward
depression, anxiety, and stress. Higher levels of stress, anxiety, and depression were
found in students living with friends as compared to those living with the test results
reflected that students with literate parents had comparatively fewer symptoms of
47
depression, anxiety, and stress than students having illiterate parents. However, no
correlation was found between these factors and family type (nuclear or joint family).
The type of accommodation also has a great impact on levels of depression, anxiety,
and stress. [46]
Viet Anh Nguyen (2017), in “The Impact of online learning activities on student
learning outcome in blended learning course”, explores the influence of learning
activities on the learning outcome of students. Student groups were examined by
involving them in interactive activities based on four major categories teacher-
student, student-student, student-content, and student-technology. A blended learning
course was introduced to students and a formative evaluation was carried out.
The experimental results were obtained by probing into learning analytics data that
was gathered from the learning management system. Quantitative data was collected
and by applying regression analysis it was found that students who participated in
interactive activities through blended learning courses gained excellent learning skills.
Thus, the author advises teachers to create activity-based learning courses and execute
them through blended learning. [47]
Rianne Conijn et.al. (2017), in “Predicting Student Performance from LMS Data: A
Comparison of 17 Blended Courses Using Moodle LMS”, studied the data of students
through the Learning Management System Moodle for the various blended course to
anticipate the progress of students during the academic year. Predictor variables were
used from LMS to analyze the attainment. It was found that for early mediation or
during evaluation, the LMS factors did not prove to be much beneficial. The author
provided a theoretical framework and the accuracy of predictive models was between
8% to 37% for distinct courses. Linear and Binary Logistic regression was used for
finding early adjustment factors. A significant improvement was reflected in the use
of LMS towards the end of the semester and before the examinations. [48]
Ryan Dale B. Elnar (2018), in “Utility Estimates of a needs-based model instrument

in College Students having academic Issues: Initial Reports of Validity and its
implications to Psychoeducational assessment and intervention”, finds the
applicability of psychological instruments in relevance to students’ personal and
academic issues. The initial validity of the model is measured by considering the
exploratory and confirmatory factor analysis.
48
The 32-item instrument is developed and the investigation is carried out using a semi-
qualitative and quantitative approach. The result proved that the SFSN-C helps to
determine the issues of students. The author concludes that the instrument will give
better insight to counselors for identifying the academic problems of students but
recommends further comparison between clinical and non-clinical results. [49]
Amit Kumar Bishnoi (2018), in “Applications of Deep Learning and Machine-

Learning”, highlighted the predominant areas, sub-domains, and applications of the
machine –learning. The author described in detail the upcoming fields where there are
chances that machine-learning would be used on a large scale. Computer vision,
prediction, semantic analysis, tongue processing, and information retrieval were the
various areas where machine-learning was applied broadly. The necessity of artificial
learning and profound learning was discussed in this paper.
The paper explained the applications of machine-learning and technical development

that scholars have been examining for the past few decades. According to the
research, there are numerous platforms available for the creation of any program of
deep or machine-learning. There are numerous new sectors where these technologies
are used nowadays. In addition, deep learning implementations have plenty of room
for expansion. In the opinion of the researcher, one can now study the newest fields of
deep learning through framework analysis. It was concluded through the research that
there is a lot of potential for developing modern tools using deep learning
architecture. [50]
Charvi Jain et al. (2018), in “Emotion Detection and Characterization using Facial
features”, worked on facial features extracted from images. The trained data was
passed through several filters like the Gabor filter, Discrete Wavelet transforms, and
Histogram of Oriented Gradients. A support vector machine classifier was then
applied to this filtered data. The classification resulted in the identification of certain
emotions such as happiness, sadness, anger, fear, neutrality, and disgust. The best
results were acquired using the HOG filter and SVM classifier with an accuracy of
85%. Viola Jones algorithm was used to detect a face. Only two features eyes and
mouth were considered expression recognition. [51]
49
Zhanjia Zhang et al. (2018), in “A Systematic Review of the Relationship between
Physical Activity and Happiness”, reviewed the research literature from research
papers that were written after 1980. Research papers related to the observational
studies, cross-sectional studies and intermediating studies were analyzed. All the
studies claimed a positive effect of exercise on the factor of happiness. It was found
that the basic and daily exercises help to increase the joyfulness of the individual. In
some papers, the study was undertaken for the senior citizens and cancer survivors to
find the level of pleasantness they experience. [52]
Rekha Nagar et. al. (2019), in “A literature survey on Machine-Learning

Algorithms”, elaborated on the use of machine-learning tools. The paper reviewed in
detail all the machine-learning algorithms. The researcher made a comparative
analysis of the traditional models with the new models implemented using deep
learning algorithms. The researcher through a literature review briefed on the popular
tools used in machine-learning.
In addition, the various libraries used in machine-learning tools were discussed. The
paper focused on how machine-learning algorithms help in predicting supply chain
demand and supply ratio, stock price, tourism, and solar irradiation. The appropriate
algorithm for each sub-domain has been explained. The paper summarizes the
application of almost all machine-learning algorithms. [53]
Kirsti Grotan et. al. (2019), in “Mental health, academic self-efficacy and study
progress among college students-The SHoT study Norway”, throws light on the
mental disturbances prevalent amongst college students and their reparations on
academic self-efficacy and student progress. HSCL-Hopkins Symptom checklist was
used to measure mental distress and the GSE-General Self-efficacy scale was
modified to measure academic self-efficacy. Apart from measuring the above two
parameters, the objective was also to find the students who took the help of
counselors and those who are thinking to get assistance. The results found by applying
logistic regression reported that the students who had major mental issues and less
counseling reflected poor academic self-efficacy as compared to those who had less
distress, early identification of symptoms like anxiety, depression, or proper guidance
from counselors.
50
The author concludes that early detection of mental health issues and mediation helps
in improving self-efficacy as well as student progress. The author puts forward the
need to assess students’ mental health parameters that are associated with academics
before the commencement of the course. [54]
Susmita Ray (2019), in “A Quick Review of Machine-Learning Algorithms”, the

most popular machine-learning techniques were reviewed in this study. This paper
will help researchers to tackle classification, regression, and clustering problems using
machine-learning algorithms. The research paper highlighted the benefits and
drawbacks of various algorithms in conjunction with using comparisons wherever
possible in terms of effectiveness, rate of learning, etc. In addition to this, the
applications of these algorithms in the real world were also discussed. It was
anticipated that it would provide information to the readers to help them make an
informed choice in recognizing and exploring the machine-learning options that are
available. [55]
Anjali Jagwani (2019), in “A Review of Machine-Learning in education”,

highlighted the applications of machine-learning and artificial intelligence in
education. The paper stated the basic classification of machine-learning algorithms.
The researcher claimed that with the advent of these techniques, educators would be
able to overcome the learning gaps between teachers and learners. The ML tools will
also help to organize student data properly. [56]
Wisal Hashim Abdulsalam et al. (2019), in “Facial Emotion Recognition from

videos using deep convolutional neural networks”, utilized a deep convolutional
neural network for emotion recognition from the facial analysis. Amsterdam dynamic
facial expression Set-Bath intensity variations (ADFES-BIV) dataset was used. The
trained model could classify ten emotions including neutral emotions. The Warsaw
Set of emotional facial expression pictures (WSEFEP), containing two hundred ten,
photographs of facial expressions were used for testing the model. The model
achieved high accuracy for all the emotions. [57]
Darshana Anil Chopade et al. (2019), in “A Survey: Django framework”, described

in detail the advantages, disadvantages, and functionalities of the Django framework.
A generalized workflow was provided that could be practiced to develop applications.
51
It was concluded by the researchers that the benefits of Django exceed the drawbacks.
Django was framed keeping in mind the beginners. In addition, it is written in very
easy-to-learn python language. It was also observed that the administrative platform
presented by the Django application is very well organized. [58]
Renieri Correia et al. (2019), in “Detecting Design violations in Django-based Web

applications”, introduced MTV checker, an automated tool to notify design violations
occurring in web applications designed using Django. It was revealed by the study
that five violations appeared frequently. Improper use of the Manager, Meddling
View, and Meddling Model accounted for about 95% of all the violations that were
found in each release. The laborious persistence method and the brain persistence
method were the two violations that declined during SUAP evolution and showed
extremely few instances. [59]
Dhwani Mehta et al. (2019), in “Recognition of Emotion Intensities Using Machine-

Learning Algorithms: A Comparative Study”, applied different algorithms to classify
and recognize the intensities of emotions. Local Binary Patterns, Histogram of
Oriented Gradients, and Gabor filters were the feature extraction algorithms used in
the study. LBP algorithm resulted in the highest accuracy. For categorizing emotions
SVM, KNN, and Random forest algorithms were used. The study was conducted to
find the best combination of feature extraction and emotion classifier algorithms,
which could be used in the future for behavioral analysis. The study was conducted on
static images stored in the database, as well as dynamic images captured through live
video streaming. Action units are widely used to measure the intensities of emotions
reflected on the face. However, according to the researcher action units proved to be a
non-reliable parameter of impulsive responses. The author suggested merging layers
within a CNN network and creating a model to address the problem of action units.
[60]
Ning Yan (2019), in “Online learning behavior analysis based on machine-learning”,

has applied ANN, to predict the correlation between the academic marks of the
students and their behavior when they are taking lessons on online platform. Different
types of behavior such as the login pattern, resource utilization, communication with
peers and teachers were observed. Some relevant parameters like number of days the
52
student attended online classes, the hit count, connection time, gender and age of the
student were considered while monitoring the behavior. [61]
Timon Elmer et. al. (2020), in “Students’ under lockdown: Comparisons of students’
social networks and mental health before and during the COVID-19 crisis in
Switzerland”, compare the undergraduate students who encountered the pandemic
during their academic course with those students who did not experience the
pandemic during their graduation period. An effort was made to capture the effect of
the pandemic on the changes adopted by students in establishing social networks, also
an attempt was made to monitor symptoms of anxiety, depression, stress, and
loneliness.
The author emphasizes the advent of a collaborative platform for students so that they
will get a better opportunity to interact with peers. This study reported that female
students show more symptoms of stress, anxiety, and depression compared. It is
reported through this paper that group study, friendship, and social support are
important for the mental health of the students, also if personal contact is not possible,
we need to find ways that will negate the effects of COVID-pandemic on the social
life of the students. [62]
Chrysi Rapanta et al. (2020), in “Online University teaching during and after the
COVID-19 crisis: Refocusing teacher presence and learning activity”, enforced the
need to plan learning activities with certain parameters, which could prove beneficial
in these changing times. According to the authors, the important aspects of teaching
roles such as social, intellectual, and coordinator should be combined properly so that
the teachers could adapt to the new and uncertain teaching-learning environments
easily. A primary requirement of online instruction is to have a specific level of
pedagogical content knowledge. For improved learning experiences and distinctive
learning environments, it is necessary to obtain a basic knowledge of handling digital
devices. [63]
Stephan Gerhard Huber et.al.(2020), in “COVID-19 and schooling: evaluation,

assessment, and accountability in times of crises-reacting quickly to explore key
issues for policy, practice, and research with the school barometer” has framed a
survey for capturing the responses of students, teachers, school management and
53
parents in line with the COVID-19 effects on education system. An effort is made to
reveal the upcoming policies, practices, and further research in this area.
The survey was carried out in schools in Germany, Austria, and Switzerland. Certain
interpretations were put forward by the author like Self-disciplined students who
manage to study for long hours also it was found that teachers’ persuasion and control
were effective in better learning outcomes. It was found that digital infrastructure
availability was less in Germany as compared to Austria and Switzerland. The stress
of the COVID-19 pandemic was seen more among students and Policymakers as
compared to teachers and parents. [64]
Ida Faridahet.al. (2020), in “Effect of Digital Learning on Student Motivation during

COVID-19”, tried to explore the desire of students to use digital learning tools during
the COVID-19 pandemic. The author concludes that digital learning has a positive
impact on the amount of student motivation in online learning. After analyzing the
student data, it was found that students were concerned regarding the cost of the
digital tools. Thus, the students expressed a need for proper cost-effective
mechanisms on the campus. [65]
Joko Joko et.al. (2020), in “Effectiveness of Mobile Learning implementation in

increasing student competence and preventing the spread and impact of COVID-19”,
attempts to evaluate some major factors firstly the extent to which disciplinary
measures were followed by students to prevent the spread of COVID-19, Secondly
examine the mentality of students who had symptoms of COVID-19 or those who got
infected by COVID-19, Impact of M-learning usage on the competence of students
during COVID-19.
A quasi-experimental method was used and it was found that many supporting factors
counted for enhancing the competence of students using M-learning tools such as
collaborative learning platform, parent support, internet connection, availability of
recorded audio, and videos due to which flexibility was achieved in the learning
process. The results reflected a tremendous difference of 0.49 in the average
competence of students before and after M-learning usage.
The author also highlights certain shortcomings of M-learning such as students with
low comprehensive skills for the English language and those with less exposure to M-
learning tools finding it difficult to adapt to the new learning patterns. The author
54
concludes that the results of this study will provide valuable insights to develop a
good learning model, which can turn out to be fruitful in adverse situations. [66]
Yizhuo Chen (2020), in “The influence of COVID-19 depression and hope on

subjective well-being in Chinese adolescents”, tried to analyze the subjective well-
being of Chinese students considering the COVID-19 pandemic. Certain prominent
correlations were uncovered by the study such as a significant correlation exhibited
between hope and the subjective well-being of Chinese students with depressive
symptoms reflected lower scores for subjective well-being. Abiding by Public health
regulations has contributed positively to the well-being of students.
The author emphasizes the need to carry out this study on a larger sample population
to get results that are more accurate. [67]
Maria Chierichetti (2020), in “Understanding the role that non-academic factors

play on students’ experience during the COVID-19 pandemic”, captured the influence
of non-academic parameters, on the psychological and overall well-being of students
during the COVID pandemic. Through the results, it was interpreted that poor
financial conditions, lack of proper technical support, uncertainty in physical health
conditions, and increase in anxiety and depression leading to poor mental health were
amongst the many non-academic factors that contribute to the memory, performance,
and confidence of students. [68]
Viet Duong et.al.(2020), in “The Ivory Tower Lost: How College Students Respond
Differently than the General Public to the COVID-19 Pandemic”, compared the social
impact of COVID -19 on University students and the general population using
sentiment analysis. The views were collected through Twitter and the author
concluded that positive tweets were less in both students and the general population
but college students display tweets that are more negative. [69]
Ashley L Ruba et.al. (2020), in “ Children’s emotion inferences from masked faces:
Implications for social interactions during COVID-19”, studied to what extent
emotions of children could be found through facial expression when certain parts of
the face like the mouth and eyes were covered with mask and sunglasses respectively
during the COVID pandemic. The experiment was carried out on schoolchildren
between the age group of 7-13 years. A racially varied sample was used for the study.
55
Random Image Structure Evolution Paradigm was used to analyze and store images
with expressions of happiness, anger, fear, and sadness. Results showed that due to
masks and sunglasses, certain emotions like anger and sadness, fear, and surprise
could not be predicted accurately. Some emotions could not be accurately predicted
but overall it was concluded that emotions could still be very well anticipated. [70]
Nurbaity Sabri et.al. (2020), in “ Student Emotion Estimation Based on Facial

Application in E-Learning during COVID-19 Pandemic”, came up with a facial
recognition application that could help teachers recognize the emotional state of
students during online learning. The facial expressions were analyzed through which
the emotional states of the students were derived.
Feature classification was done through Support Vector Machine and Support Vector
Regression. Support vector regression showed better results. SVR reflected optimistic
results and the application was successful in recognizing four basic types of emotions
Happy, sad, normal, and surprise. Results showed 99.16 accuracy percent from SVR,
whereas SVM reflected 33.33 percent accuracy. [71]
Adamya Shyam et.al.(2020), in “ A Django-based educational resource sharing

website: Shreic”, developed a resource-sharing website named “Shreic” that enabled
students to access various academic facilities such as video lectures, subject notes,
PowerPoint presentations, syllabi of various courses and old question papers. The
module for selling old books was also included, which helped the students.
The project was designed using Django and a software development lifecycle iterative
approach was used for implementation. Python and Django were used at the backend
along with Jinja2, python engine, and SQLite, which is a C-language library. The
researcher concluded that the project proved to be beneficial for students as they
could get many facilities through this online platform. According to the author, the
Systematic approach of SDLC should be used for the development of other web
applications. [72]
Anuradha Khattar et.al. (2020), in “ Effects of the Disastrous Pandemic COVID-19

on Learning Styles, Activities, and Mental Health of Young Indian Students- A
Machine-Learning Approach”, discussed how the pandemic brought about changes in
the daily routine of the students, in their learning patterns and overall behavior.
Students in the age group of sixteen to twenty-three responded to an online survey and
56
expressed their feelings about online learning, digitization, social connectivity, and
mood swings during the lockdown. The data collected was then examined using R
programming and Python. Graphical analysis was done through MS Excel. The Prior
algorithm was used to understand associations considering thirteen parameters.
Association rules were assessed using Support, Confidence, and Lift as measures of
interestingness. The experimental results reported that uncertainty due to pandemics
leads to a mental and emotional disturbance among students. The responses from
students affirmed that online education and technical tools could be used during the
pandemic and in the future to assist teaching learning but online teaching learning
could never be a replacement for the offline education system. [73]
Lijia Chen et al. (2020), in “Artificial Intelligence in Education: A Review”, studied

the influence of Artificial Intelligence on schooling. The scope of the study was
restricted to AI’s use and impact on management, instruction, and learning. The work
was realized through the employment of a qualitative research approach. As AI,
evolved teachers were able to review and grade student assignments more effectively
which lead to quality assessment. Machine-learning algorithms were able to add the
factor of personalization that helped educational institutions to deliver a better
learning experience to students. The major focus of the paper was to find the effect of
AI on administration, teaching learning, and evaluation. [74]
Katerina Kedraka et al. (2020), “Effects of the COVID-19 Pandemic on University

Pedagogy: Students’ Experiences and Considerations”, surveyed Greek students to
understand their feelings regarding distance learning in the initial phase of the
lockdown. According to the responses given by students, it was interpreted that online
learning was favorable for them, but they were missing the practical training, which
was very much essential in subjects like molecular biology. Thus, the need for
modifying the educational equipment expressed by the researcher. [75]
Andi WahyuIrawan et al. (2020), “Psychological Impacts of students on Online

Learning during the Pandemic COVID-19”, conducted a telephonic interview to learn
the mentality of students towards the sudden shift to online classes. A qualitative
technique was used and it was found that students in the lower-income group were
anxious, as they had to arrange and invest for resources to continue their education.
57
The students experienced mood swings. In addition, it was found that there is a need
to increase the psychological counseling of students for their welfare. [76]
Dr. Pravat Kumar Jena (2020), in “Impact of Pandemic COVID-19 on Education in

India”, highlighted the Indian government's initiatives taken in India for facilitating
education during COVID pandemic. The researcher prominently mentioned the
positive and negative factors and changes associated with the pandemic concerning
academics. The author suggested that for practical subjects online teaching proved to
be beneficial. India should develop and come up with a strong platform for online
learning, which will then be an alternate and reliable source for imparting knowledge.
[77]
Toshika Pareek et al. (2020), in “A Comprehensive Study on COVID-19 Pandemic:

An Impact on School Education in India”, observed certain facts about the online
education scenario in India during the pandemic. The view of students, teachers,
parents, and administrators was studied through structured questionnaires. The impact
of the pandemic was also examined by the gender of the students. Around one
hundred respondents from public and private schools of Udaipur city of Rajasthan
were considered as the sample for the experimental survey. It was summarized that
traditional classrooms should be updated with technical equipment, as during
COVID-19 students were used to online platforms. In addition, the study expressed
the need to train teachers, parents, and administrators in the better use of ICT tools.
[78]
Seema Patil et al. (2020), in “ICT plays a vital role in social and educational sector
on accidental occurrence of COVID 19”, discussed the features, applications, and
components of ICT. Applications of ICT in the social and educational sectors were
discussed. With the emergence of the pandemic, the ICT tools like video
conferencing, the arogya setu app, drones, and similar other devices were used which
were discussed in the paper. The researcher summarized the need to increase the
awareness and skills of students and teachers equally to adapt and handle ICT-based
education effectively. [79]
Ashish chandiramani et al. (2021), in “Management of Django Web Development

in Python”, explained how one could create interactive and user-friendly interfaces
58
using tools and commands of Django web frames. Back-end language was also
studied and was found to be used earlier only for specific purposes. However,
nowadays it has emerged as a topmost language for website development, Gaming,
AI, and advanced mobile applications. A Weather application was developed using an
administrative panel. The application could monitor the weather conditions and
forecasts of various cities. [80]
F.M Javed Mehedi Shamrat et al. (2021), in “Human Face Recognition applying
Haar Cascade Classifier”, put forward a framework for PC- a based face recognition
problem. Frontal view of the face is vital in face recognition but there are many
situations in which one could not capture the frontal features appropriately. The
researcher highlighted the need to explore the research of imperfect images that are
cached using CCTV cameras. Training the dataset consisting of Partial frontal
features is still challenging and the researcher has made efforts to explore this topic in
this paper.
The proposed system operated in three major parts facial data gathering, training the
stored image, and face recognition using LBPH. In the dynamic Cascading method,
face detection and analysis are done by selecting Haar features. An integral image is
generated on which cascading classifiers are applied. In the proposed structure the
Local binary patterns histogram estimate is utilized to create a dataset. The final
computational component is achieved using image processing. Each individual will be
identified using an Id number. During the identification process when the image of the
test individual matches the image in the dataset, the machine recognizes the face.
The researcher has taken 200 photos of each person and clustered them under the
same id. Pictures were captured from real-time videos also and cropped image of
150*150 pixels was saved in the database with a distinct id.’yml’ file was created
storing matrices of every picture. Python programming language was used for
experimenting. Two tests were conducted for analyzing the results of the experiment.
In test 1, only the faces in the database were tested. While in test 2, the same faces
with different external settings were tested. [81]
Rouvrais Siegfried et al.(2021), in “Preparing 5.0 engineering students for an

unpredictable Post-COVID World”, revealed approaches to groom students and make
them industry-ready, especially for the future work environments which are expected
59
to be generated as a result of the COVID-19 pandemic. Through the study conducted
on European and technical Universities, the author concluded that academicians and
industry experts should focus on bridging the gap between industry and academia; for
this, certain measures shall be taken like restructuring the syllabus, improvising
technical skills of students, and enhancing entrepreneurship skills of students. Thus,
more employable students or entrepreneurs could be created and students could
become capable enough to face future uncertainties if they ever occur in upcoming
years. [82]
Peter Kolar et.al.(2021), in “Experiences with online Education during the COVID-
19 Pandemic-Stricken Semester”, highlighted the effect of online education on the
final result of students who experienced a sudden shift from offline to online
academic education for the first time due to COVID -19 pandemic. The final grades
obtained by engineering college students were analyzed and it was observed that the
number of lecture views increased drastically toward the end of the semester. The
author concluded that there is a need for rigorous study for finding a tool that could
monitor the association between the students’ study habits and the outcome of the
attended online course. Two major recommendations are given by the author firstly
the duration of videos should be less and students shall be frequently assessed which
may probably lower the last minute cramming of students and will lead to better
online course delivery. [83]
Nabil Hasan Al-Kumaim et.al. (2021), in “Exploring the Impact of the COVID-19
Pandemic on University Students’ Learning Life: An integrated Conceptual
Motivational Model for Sustainable and Healthy Online Learning” studied the
challenges faced by University students while adapting the online education and the
technical platform. The author concluded that family support, University support, and
teacher and student training on better educational platforms would help in fruitful
online learning. [84]
Michal Wilczewski et. al. (2021), in “The Psychological and Academic Effects of
Studying from the Home and Host Country during the COVID-19 Pandemic”, put
forward the analysis of international students through a survey carried out in Poland.
The major research findings reflected that the students who returned to their home
country and took up online classes found discussion with peers as a helpful factor in
60
achieving pace with online education. A prominent increase was observed in students’
loneliness factors in both home and host countries. [85]
Galina Ilieva et.al. (2021), in “Effects of COVID-19 Pandemic on University

Students’ learning”, designed a new framework to evaluate the opinion of students on
online education due to the pandemic. Machine-learning algorithms like SVM, RF,
CTREE, and CART were used to classify the data. The author concluded through
experimental observations that students with lower income groups encountered more
challenges towards online education as compared to students who were in better
financial conditions as they were technically well equipped. In addition, it was found
that students with poor living and financial conditions and those who spend less time
studying online were at high risk of discontinuing education. [86]
Eduardo Puraivan et. al.(2021), in “Emotion-based decision support tool for

learning processes An application with undergraduate students during COVID-19
pandemic”, created a decision support model to analyze the emotions of students that
will help teachers to conduct online teaching more efficiently. Positive and Negative
Affect Schedules (PANAS scale) and Natural language processing techniques were
used to capture the emotions of students. The author concluded that the tool would be
beneficial for capturing important factors of students like level of motivation, and
difficulties faced by students. The tool can assist the teacher in creating project teams
and career counseling. [87]
Vartika Bhadana et. al.(2021), in “ A Comparative Study of Machine-learning

Models for COVID-19 Prediction in India”, came up with a framework using
Machine-learning algorithms to predict the active cases, death count recovered
patients, the number of persons who have a chance to get infected in the upcoming
five days. The linear regression and LASSO model represented the best results
compared to other machine-learning models. SVM was proved the worst predictor.
[88]
Cheng-Sheng Yu et.al. (2021), in “A COVID-19 Pandemic Artificial Intelligence-

Based System with Deep Learning Forecasting and Automatic Statistical Data
Acquisition: Development and Implementation Study”, developed the COVID -19
pandemic AI system to find the variations, mortality rate, and future trends for a
61
COVID-19 pandemic. Deep learning algorithms are used to predict worst-case
scenarios. The Model also has updates on vaccination, which can change the
pandemic situation. [89]
Uriel Alan Flores-Juarez et al. (2021), in “Two Facial Emotion Detection Based on
Naïve Bayesian Classifier”, implemented the naïve bayes algorithm to generate
emotion classifier model. The model divides emotions into two major classes happy
and sad. The distance between the nose to the lower lip, nose-eyelid, and between
nose and eyebrow were computed. These features were selected as the emotions were
most strongly represented by them. According to the researcher, the subjective
parameters fed to the system will give a rough estimate of the emotions. The study
proposed a robust system with an innovative interface that could easily test the
performance of the algorithms and thereby help in increasing accuracy. [90]
Sanna Ruhalahti et al. (2021), “Identifying Higher education First- Year Students’
Reported Experiences Studying during the Pandemic”, observed first-year students of
a bachelor’s degree course who had taken admitted in January 2021. Some interesting
patterns were uncovered by the researcher. As it was the second phase of the
pandemic lectures were conducted online. According to the researcher, students who
could smartly handle technological equipment and were having enough resources
reflected a positive approach toward online education, especially during their initial
year of University education. Those who lacked self-motivation, and had limited
resources showed a negative approach toward this type of education. The survey was
conducted on One hundred fifty-four Finnish students. The inductive method was
used for result analysis. The survey results emphasized the need to strengthen student
counseling and build better technical and pedagogical competencies of teachers. [91]
Jianmei Zhang et al. (2021), in “Rural-Urban disparities in knowledge, behaviors,

and mental health during COVID-19 pandemic”, investigated certain parameters like
knowledge, behavior, and state of mind of people living in city and rural areas. The
results were compared and it was found that people living in urban areas were very
well aware of the symptoms, precautions, and care related to COVID as compared to
people in rural areas. However, the anxiety and depression levels were more in urban
residents in comparison to rural residents. [92]
62
Rana Saeed Al-Maroof et al.(2021), in “The Effectiveness of Online Platforms after
the Pandemic: Will Face-to-Face Classes affect Students’ Perception of their
Behavioral Insights Unit (BIU) to use Online Platforms”, tried to find the desire of
students to use online tools and equipment post-pandemic during offline classes. A
hypothetical model was framed considering seven attributes situation awareness of the
user, ease of use, the usefulness of the tool, quality of education system, Information
quality, Richness of Information, and satisfaction of use. Information richness was
influenced by situation awareness. Whereas, user satisfaction was dependent on the
ease of use and usefulness of the tool. The results show that IR, Education system
quality, information quality, satisfaction, and TAM were key factors in evaluating the
probability of students using the technological tools. [93]
Galina Timokhina et.al.(2021), in “Digital Transformation of Educational services

in COVID-19 Pandemic: Social and Technological Aspects”, focused on social and
technological changes in the educational sector that emerged to the sudden transition
to online and distance education due to the COVID-pandemic.
The research is dedicated to three major objectives. In the first section, major
literature and secondary data have been analyzed to find the effect of using online
platforms in the education sector and to determine the social and technical
implications of the same. The author analyzed the major contribution of the
Government in setting up high-speed internet connections, expansion and use of
wireless access zones, and usage of digital devices and their integration into
academics.
There was an increase in the use of free and international Wi-Fi service “eduroam”.
New “eduroam” access zones were established. In the second section, the author
enlisted some major LMS like Moodle, and Red class, and found that certain video
conferencing tools like Zoom and Microsoft teams gained popularity. The statistical
results showed that there was a 546% growth in the use of Presentation tools, the use
of interactive tools raised to 600%, whereas automation tools increased by 1,310%.
The author analyzed that the investment of the Government in integrating online
platforms and digital devices in academics determined its use. In the third section,
students of certain Russian Universities were surveyed to find their contentment with
educational services. In addition, the teacher’s perspective on online teaching was
63
examined. Results reflected that around 64% of the students were happy with the
organizational efforts that were inculcated to shift to the digital platform.
The major concern was the lack of infrastructure, power supply, and financial
resources to acquire these facilities. Around 65% of the students expressed that online
learning was less effective than traditional learning. Lack of teacher training programs
leading to inadequate use of digital resources, long hours of lesson preparation time,
and difficulties in conducting hands-on sessions were some of the difficulties faced by
the teacher community.
The author concluded that it is difficult to predict accurately the effect of a pandemic
on the social, technological, and financial aspects of online education. The author
further expresses the need to conduct detailed research to come out with a well-
organized education system that can sustain in pandemics and times of uncertainty in
the future. [94]
Iman Akour et.al. (2021), in “Using Machine-Learning Algorithms to predict

people’s intention to use Mobile Learning Platforms During the COVID-19
Pandemic: Machine-Learning Approach”, highlighted the use of mobile learning
applications in the process of teaching-learning in the context of COVID-19. The
technology acceptance model (TAM) and theory of planned behavior (TPB) these two
models were used by the author to understand the acceptance of technology by
educators and students.
The researchers designed a hybrid model to include external factors that are subjective
Norms and Fear. The proposed model checked the influence of SN and Fear on the
Perceived ease of use (PEOU) and Perceived usefulness (PU) of mobiles. The result
of the study was examined by two techniques the first one being the smart PLS tool
and the second one being the machine-learning algorithms. The behavior of the
students in the classroom and daily routine activities was predicted. Fear, attitude,
apprehension, and anxiety were also monitored concerning the willingness of students
to use mobile applications.
BayesNet, logistic regression, and J48 classifier were some of the mobile learning
algorithms that were used. It was interpreted that the J48 classifier depicted 89.37%
accuracy and performed better than other algorithms. It was concluded by the author
that the use of mobile applications would help achieve betterment in education; On
the other hand, the emotional imbalance caused by the pandemic could reduce the
64
value of such systems. Thus according to the author, it is important to monitor the
emotional and behavioral state of the students to upgrade digital usage and its
effectiveness in academics. [95]
Tom Crick et.al. (2021), in “The International Impact of COVID-19 and “Emergency
Remote Teaching” on Computer Science Education Practitioners”, compiled the
changes imposed by the COVID-19 pandemic in the field of education specifically in
Computer Science. International Survey was carried out and it was found that teachers
in the field of computer science found it easy to accept the changes in teaching,
learning, and evaluation as compared to teachers of other streams.
High school teachers could easily adapt to this transition as compared to Teachers in
Senior Colleges or Universities. Imparting Practical Subject Knowledge to students
was the only common issue that was faced by teachers at all levels. The Chi-Square
test was applied to the dataset. The author expressed the need to upgrade and find
some universal approaches to teaching basic Computer science concepts globally with
the help of an online platform that could be useful in the future. [96]
Jing chang et.al. (2021), in “Online Instruction and Offline Classroom teaching: a
study on Parallel Education Systems highlights the benefits of online teaching-
learning. Through the acquired results the researcher suggested that effort should be
made to imbibe online teaching especially to improve the pre-class, in-class, and post-
class attainments of students. In the offline teaching method, the body language of
students helps teachers to understand if the students got the concepts taught; but in
online mode, facial expressions and interaction with students are somewhat less.
Hence, according to the author, there is a need to find online education tools and
platforms, which could track and assist the attainments of students before teaching,
during, and after teaching. There is a need to find and include the best features of
online structure in offline learning which could be helpful in the long term. [97]
El-Sayed Atlam et.al. (2021), in “A new approach in identifying the psychological

impact of COVID-19 on University students’ academic performance”, explored the
effects of COVID -19 on college students. The statistical and machine-learning
approaches were implied to the data to find the psychological influence on the
academic performance of students. The results depicted a significant correlation
between mental state and acceptance of online education.
65
Similarly, there was a positive relationship between digital tools used to impart online
teaching and the academic performance of students. The effect of long-term digital
tool usage on the academic performance of students residing in different geographical
locations was evaluated. Machine-learning models like Logistic Regression, Decision
tree, SVM, and deep learning reflected precise results. The lowest performance was
reported by the AdaBoost method. Parents as well as educational authorities faced
some negative effects such as the pressure of availing online education students. On
the other hand, lack of understanding of practical subject concepts, and improper use
of technology were some of the factors that diverted students from online education
and proper academic achievements. [98]
Ankan Bhattacharyya et.al. (2021), in “ A deep learning model for classifying

human facial expressions from infrared thermal images”, presents an experimental
study to infer facial expressions through a deep learning model called an infrared
facial expression network. Two major modules residual unit and transformation unit
are used. The snapshot ensemble technique was used to capture thermal images for
better precision and efficiency. The deep learning model proposed by the researcher
shows 88.43% accuracy. Infrared thermal cameras were used for capturing images as
these cameras work well in dim light also. Emotions like happiness, sadness,
neutrality, surprise, fear, anger, contentment, and disgust could be identified through
the model. The deep CNN model is used for feature extraction. [99]
Arijit Nandi et. al. (2021), in “Real-time Emotion Classification using EEG data
stream in E-learning Contexts”, designed a real-time emotion classification system
using a stochastic gradient descent algorithm. The model is trained for online
response collection from students to capture emotions during the online streaming of
classes using electroencephalography (EEG). Logistic regression was used for
training the model. The F1 score for the proposed model is better than other online
and offline methods. The model is useful in identifying emotions during online
streaming efficiently which is very much useful in the pandemic situation when
education shifted online. [100]
Mohammad Nehal Hasnine et.al. (2021), in “Students’ emotion extraction and

visualization for engagement detection in Online Learning”, developed an educational
application to find the emotions of students during online classes thereby finding the
66
level of students' interest in academics. The computer vision method was used to
extract the emotions of students and a trained Convolutional Neural Network was
used to classify students depending on their involvement in online lessons. The study
would help teachers’ group students for project activities and inquiry-based learning
while considering the emotions of students. [101]
Taoufik Ben Abdallah et.al. (2021), in “Student Behavior Recognition in Classroom

using Deep Transfer Learning with VGG-16”, built a framework using a deep
learning algorithm. There are three major steps first face analysis, second face
expression recognition, and finally behavior classification of students. Haar Cascade
algorithm was used to detect faces. Facial features were abstracted using Visual
Group Geometry (VGG-16). VGG-16 was used in the experimental study. Python
Keras library with TensorFlow at the backend was used to identify and categorize
students’ behavior.
The accuracy of the results was improved using the data augmentation technique.
Further three neural network architectures with VGG-19, VGG-16, and VGG-16
Modified were compared to find the best outcomes. Results predicted 68.3% accuracy
for modified VGG-16. The study was helpful to estimate the understanding and
behavior of students on the online platform. [102]
Bo Yang et.al. (2021), in “Face Mask aware Robust Facial Expression Recognition
during the COVID-19 Pandemic”, used a binary deep classifier and trained facial
expression recognition tool to generate a system that could classify masked and
unmasked parts of the face and could further help in detecting accurate emotional
readings through facial expressions. Attention weight maps and Reverse attention
weight maps were derived to analyze covered and uncovered facial parts. The design
suggested for Facial expression recognition stressed more on capturing unmasked
regions as compared to the masked parts of the face. The model showed good results
and predicted positive, either negative, or neutral emotions. [103]
Akshi Kumar (2021), in “Machine-learning for psychological disorder prediction in

Indians during COVID-19 nationwide lockdown”, highlighted anxiety disorder and
mood disorders, which are the two primary categories of psychological disorders.
Through a mental health questionnaire, 395 Indians were surveyed to detect the
psychological load developed. A psychological disorder prediction tool was proposed
67
which helped in identifying the initial stages of the disorder. A multinomial Naïve
Bayes classifier was used to instruct the model. The experimental results gave 92.15%
accuracy in classification. The researcher concluded that the results would be
beneficial for early medical interventions. [104]
Rakesh Kumar Singh et.al. (2021), in “Django Web Development Simple and Fast”,
explained in detail the Django framework with few extensions. The author illustrates
how the Django framework is better compared to other designs available. Installation,
initiation of a new project, and Model view template components are briefed in this
paper. The researcher provided all the basic knowledge required for creating an
application such as developing class views, function views, form design, and object-
relational mapper using this technical structure in Python. The use of Django
Templates and Django Template language was also described in this research paper
[105]
Ankit Verma et.al. (2021), in “Web application Implementation with Machine-

Learning”, created an interactive college website for students, staff, Management,
administrative staff, and alumni students. The website was developed using React JS
at the front end and Django at the back end.
The website was then deployed using the Amazon web services (AWS) cloud. Flask
was used to deploy it in web applications. Scikitlibrary was used to implement
Python. Text recognition and classification were done using machine-learning
algorithms through Natural Language Processing. The purpose of text classification
was to find profane text in the data entered by the user. The model helped to detect
profane text and deleted it if found for all the users. SVM, Naïve Bayes, and KNN
models were used and compared, out of which the Support Vector Machine (SVM)
displayed a 94.5% result, which proved to be the best. [106]
Xun Zhou et al. (2021), in “A Survey on evolutionary construction of Deep Neural

networks “, turned the automated DNNs into multilevel, multiobjective, optimization
problems with constraints. This issue was analyzed to gain a thorough understanding
and offered a detailed analysis of EA-based methods for optimizing the model
parameters, model architecture, and model fit as primary approaches to tackling this
challenge. The benefits and weaknesses of the EA-based methods were also
68
elaborated in contrast to other widely employed methods in many optimization
scenarios.
Two key elements of optimization, namely computational speedup, and optimization
objectives were discussed. Further, the researcher provides a summary of the uses,
issues, and development in this subject of study. The research work intends to assist
DNN researchers to comprehend the reasons and methods. [107]
Pramoda R et al. (2021), in “Emotion Recognition and Drowsiness Detection using

Digital Image Processing and Python”, recommended a system to identify emotions
and monitor if a person was dazed at a particular moment. An effort is put to create a
cost-effective and coherent system using artificial intelligence and Image processing
techniques. The important feature of the application is to detect the eye-blinking
status, which could further help to find if the automobile driver is feeling drowsy.
This facility claims to avoid automobile accidents.
The system could also be implemented in academic institutes to explore the emotional
state of students and their academic interests. The head position detection unit, face
detection, facial feature extraction unit, eye-moment, shut-eye detection, yawing state,
laziness, and alert unit were the important components of the system. The automation
of most of these units was done in experimental settings. Eye and mouth movements
were diagnosed using the Haar Cascade algorithm.
A Convolutional Neural network algorithm was applied for face recognition. Through
the system, the alert is sent to the registered mobile of the automobile driver if he is
found sleepy. The model was examined to find the issues related to face and
emotional approval. The accuracy acquired was also better compared to the traditional
system. [108]
Nahal Salimi et al. (2021), in “College students mental health challenges: Concerns
and Considerations in the COVID-19 Pandemic”, uncovered the extra worries that the
pandemic caused among students. Mental health issues encountered by children as
well as methods used by counsellors to handle these were discussed in the paper. It
was concluded that social help, and on-campus support, should be extended. Failing
to do this it was predicted that it might lead students to quit the academic sessions.
[109]
69
Fei Yan et.al. (2022), in “Framework for identifying and visualizing emotional
atmosphere in online learning environments in the COVID-19 Era”, developed an
opposite online learning environment (AOLE) that bears an intelligent 3D
visualization of the classroom setup. The AOLE tool was implemented on fifteen
students and three instructors and it was observed that the tool helped to acquire
insight into learners’ level of conceptual understanding, involvement, and
concentration. The tool interprets the emotional curves and tracks the individual
learning progress of the student. Interventions could be made in the teaching-learning
process during the course with the help of the tool. [110]
Hanan E. Abdelkader et.al. (2022), in “ An Efficient data mining technique for

assessing satisfaction level with online learning for higher education students during
the COVID- 19”, proposed a model for finding the academic fulfillment of students
toward online learning during the lockdown. Feature selection from education data
mining was used to determine the satisfaction of students. [111]
Yijun Zhao et.al (2022), in “Different coping patterns among US Graduate and
Undergraduate students during COVID-19 pandemic: A Machine-learning approach”,
designed a framework to find the behavioral patterns of undergraduate and
postgraduate students. Association rule mining was applied innovatively and it was
found that the students exhibited different mechanisms to stand against the distress.
[112]
Shuang Xue et.al (2022), in “Resilience and Prosocial Behavior among Chinese
University Students during COVID-19 Mitigation: Testing Mediation and Moderation
Models of Social Support” used a convenience sampling method and surveyed three
hundred thirteen University students in mainland China. The survey was conducted
online. The mediating and moderating roles of social support were tested to
understand the link between resilience and prosocial behavior.
The Hayes process macro for SPSS was used for investigating the results. It was
found in the study that the use of social assistance could mitigate the detrimental
effects of COVID-19 on university students’ prosocial conduct. The support
functions as a buffer between these two factors in extremely stressful situations.
Different facets of social help were measured in this study. It was found that the
students who were highly adaptive to the pandemic and similar adverse situations
70
actively participated in social activities and willingly provided a helping hand to
people in need.
Another aspect that was influential in the association between resilience and prosocial
activities was the level of support utility by students. The findings of this paper have
put forward significant developments for effective post-pandemic therapies aimed at
the psychosocial route connecting resilience with prosocial attitude. [113]
Feriha Fatima Khidri et.al (2022), in “Physical Activity, Dietary Habits, and
Factors Associated with Depression among Medical Students of Sindh, Pakistan,
During the COVID-19 Pandemic”, conducted an online cross-sectional study to
examine the dietary patterns, physical activities, and sleeping habits of students. An
online survey was supervised to find out how the above factors determine the level of
depression in students. Three structured questionnaires were given to the students for
getting the details related to their location and information related to the factors of the
study.
The outcomes showed that 28% of the medical students showed mild depression, 22%
were found with moderate depression, 9.3% with moderately severe depression, and
another 9.3% with severe depression. The pandemic restricted the physical exercise
schedule of the children and disturbed their sleeping patterns. The researcher
concluded that it is the need of the hour to encourage healthy eating habits, and a
good lifestyle to reduce mental stress. The need for lifelong planning concerning
various future transformations is required to be done. [114]
Swadha Gupta et al. (2022), in “Facial emotion recognition based real-time learner
engagement detection system in online learning context using deep learning models”,
adapted a deep learning-based approach to identify the involvement of students during
online classes through facial expressions. The system calculated an engagement index
(EI) and classified the learner into two major categories “engaged” and “Disengaged”.
Inception-V3, VGG-19, and ResNet-50 were correlated and the researchers came up
with the best predictive model for detecting real-time emotions. FER-2013, CK+, and
RAF-DB were used to evaluate the overall performance of the proposed system.
ResNet-50 was able to achieve the highest performance of 92.3%.
Inception-V3 reflected 89.11% accuracy while VGG-19 showed 90.14% accuracy.
The final model was tested on 20 online learners and it successfully identified the
71
student state as “engaged” or “disengaged”. According to the researcher, the collected
information of the proposed model could be integrated with other sensors such as
heart rate and EEG signals. In the future, a large dataset can be created to train and
test the dataset with the proposed approach. The proposed model could be applied to
children with special needs. [115]
Yuwei Chen et al. (2022), in “Deep learning based emotion detection”, tried to detect
the true emotions of the participant through computer vision, semantic recognition,
and audio feature classification. In this experiment, the traditional CNN MobileNet
model is revitalized. A new model framework ms_model_M was applied on FER-
2013 and AffectNET. These are the two frequently used real expression datasets. This
new structure has fewer parameters that were around 5% of the number of traditional
CNN MobileNet models. The new model achieved an accuracy of 74.35% and
56.67% individually, whereas the traditional model was able to achieve 74.11% and
56.48% accuracy on the respective datasets. 72% accuracy was achieved for speech
recognition. The model was trained on the Kaggle dataset. The application was able to
give voice estimation. VADER’s emotion recognition library was used to understand
text emotions. The results of voice, audio and facial expression recognition were
noted individually. These results were not integrated to evaluate the overall state of
mind of the user. [116]
Dr. M.Neelakantappa et al. (2022), in “Students live Behavior Monitoring in online

classes using artificial intelligence”, developed a platform to observe student behavior
in online classes and retain useful information about them. This study assists teachers
and will help them in decision-making. Teachers would gain a system approach by
relating classroom behavior with academic results. [117]
Ashmina Khan et al. (2022), in “A survey on Prediction and Analysis of students’

academic performance using Machine-Learning Technique”, experimented to
investigate student behavior within the classroom as well as on campus. Machine-
learning algorithms were then applied to categorize the students based on their
behavior. The classification helped to identify students who were towards the negative
behavioral pattern. Thus, the prediction model could be used to increase the positive
interaction with such students and improve their performance. [118]
72
IsmailaTemitayo Sanusi et al. (2022), in “A systematic review of teaching and
learning machine-learning in K-12 education”, conducted a review of research papers
published in conferences and journals to find out the loopholes in integrating
machine-learning in educational programs. In addition, the authors expressed the need
to include faculty development programs to provide machine-learning education to
teachers. The review also helped to find areas where further ML research is required.
[119]
Katerina Kedraka et al. (2022), “Considerations for University Pedagogy: Distance

Learning One Year after the COVID-19 Pandemic Outbreak”, collected responses
from students studying in two different Universities in Greece. The opinion of the
students was collected regarding e learning at the initial stage of the pandemic. Again,
in the next phase when the students returned to face-to-face classroom education, the
survey was taken and then the responses were compared and analyzed. According to
the study, it was found that students expressed the need to introduce and merge
technology tools in the academic system. [120]
Mrinal Bhardwaj et al. (2022), in “E-Learning: Online Learning Platform”,

supported the requirement to plan e-learning techniques and tried to switch to e-
learning tools for betterment in an online educational environment. A thorough and
critical evaluation of several earlier and more recent relevant studies was conducted to
get a better comprehension of the e-learning approaches. The paper concentrated on
the study of the biggest obstacles and challenges that were posed by stakeholders to
put e-learning technologies into practice. In the proposed work, Django was used to
design an application. The front-end view was formed using MVT architecture. The
testing of the model was done by Postman, which is used to test an API. Some
properties of the existing model are retained and the researchers in the proposed
model added new required fields. [121]
Omran Alharbi et al. (2022), in “Review theories of E-learning”, investigated the

acceptance of technology usage by teachers and learners. Further, the paper presented
a detailed review of several theories such as Social Cognitive Theory, Theory of
planned behavior, and Diffusion of innovations, Motivational Model, and Model of
Personal Computing utilization. The unified theory of acceptance and use of the
technology model was studied. The applications, advantages, and disadvantages of the
73
Technology acceptance model (TAM) were mentioned. In addition, it was concluded
by the researcher that several researchers chose and implement the technology
acceptance model (TAM) and it came out as an effective theoretical model for
learning installation and utilization of eLearning. [122]
Amjad Rehman Khan et al. (2022), in “Facial Emotion recognition using

conventional machine-learning and deep learning methods: Current achievements,
analysis, and remaining challenges”, provided an analysis of Facial emotion
recognition (FER). The review was based on several deep learning and traditional
Machine-learning methods. Publicly accessible FER datasets were also inspected with
benchmark outputs. The goal of this study was to emphasize the potential research
gap in this field. A comprehensive evaluation of FER was put forward by using
conventional ML and DL approaches. [123]
Pablo Duo Terron et al. (2022), in “ICT motivation in sixth-grade students in

pandemic times- The influence of Gender and Age”, found the student’s interest,
usage, and liability towards ICT tools. The data was collected for 924 students
between the ages of 10-13 years, which counted for around 78% of the total number
of students enrolled for class Sixth. A validated questionnaire was designed through
the ICT scale used by Programme for International Student Assessment (PISA). The
data were analyzed using the quantitative method and an experimental descriptive and
correlative method. According to the findings, the motivation and commitment to ICT
usage by students were moderate on Lickert 5 scale overall mean. Boys outperformed
girls in the use of ICT tools. This study exhibited a strong relationship between age
and gender. In addition, these two factors imposed an impact on ICT participation and
use. [124]
Raj Sharan Shahi et al. (2022), in “A 3-Year Cohort study of Internet use and
Parent-Child relationship Among Senior Secondary School students”, investigated the
growth of internet use and its influence on the parent-child relationship. The study
was limited to Indian citizens. The researcher found a correlation between the
intention of children to access the internet and related statistics on the relationship
between parent and child. The frequency of use, the preferred duration and timing, the
gender of the student, and the intent of use were the considerations for the study. The
work concluded with the need to achieve a balance between the real and virtual
74
worlds of the academicians as well as parents. In addition, policymakers should
intervene and restrict the use of the internet. [125]
Lydia T. Liu et al. (2022), in “Reimagining the machine-learning life cycle to

improve educational outcomes of students”, reviewed the existing machine-learning
methods, which are used in education. Attempt was made to find the relevance
between the existing ML approaches and their significance in updating the
educational scenario. The author has studied several research papers, which explain
the technical aspects, and ways in which machine-learning is used in betterment of the
academics.
Through the review, some loopholes were uncovered. Firstly, it was determined that
the machine-learning mechanisms that were framed were not in line with the issues
that the academicians are facing. Secondly, the results that are predicted from the ML
models should be used promptly in improvising the educational process wherever
required. In short, if early mediations are applied then at times it could turn out to be
beneficial. Thus, the researcher expresses a need to modify the existing framework.
[126]
Shijing Yuan et al. (2022), in “Effects of Physical Activity on College Students’

Subjective Well-Being during COVID-19”, studied the questionnaire responses from
students of eight different universities of Wuhan. Firstly, the student’s involvement in
physical exercise was determined. The types of physical exercise that the individual
was doing and the time spent on the physical activities were the important parameters
to decide the level of physical activeness. It was observed that those who projected
high or middle level of physical activity showed more subjective wellness as
compared to those who were less actively involved in the physical exercises. Physical
activities like sports were also found to be effective in maintaining mental and
behavioral stability. [127]
Amel Fawzi Zaki (2022), in “Investigating the Impact of Extracurricular activities on

the Academic and Social Skills of University Students in Post-COVID-19: A Case
Study”, investigated the role played by the extracurricular activities in enhancing the
social and academic skillset of the students. There are so many extracurricular
activities. The students were asked to give a write up for the magazine. This activity
75
was selected for the study. Two set of questionnaire were framed. One was for
examining the social skills and the other was for the academic skill testing.
It was noticed that the student who were the participants of the extracurricular
activities turned out to be more constructive. They could achieve the skillsets. Rather
the findings of the experiment encouraged the children to participate in more such
extracurricular activities. They felt that such type of activities are helping them to
make up for the loss that they faced in the times of the pandemic. The researcher
expresses the need to involve the students in more performance oriented
extracurricular activities, which will help them to solve issues generated because of
online learning. [128]
Ahmad Khairulanam Mohamad et al. (2022), in “Co-curriculum activities during

the COVID -19 pandemic and their implications towards soft skills”, explored the
effect of co-curricular activity participation on the grasping of soft skills. The study
focused on student participation in co-curricular activities in the times of COVID-19.
It was observed that the parents were reluctant in allowing their wards to actively
participate in the co-curricular activities. However, the schools ensured to follow the
standard operating procedures.
From the results, it was found that the summarized impact of the co-curricular activity
participation was average in the new normal. Consequently, the achievement of the
soft skills was also at a medium level. It was understood from the experimental results
that the student’s capabilities in achieving soft skill knowledge improves by
participating in the co-curricular activities. [129]
Peijun Jiang et al. (2022), in “Exploring the effects of online learning on EFL
learners’ motivation, anxiety, and attitudes during the COVID-19 pandemic: a focus
on Iran”, examined the approach of EFL (English as a foreign language) learners
towards online leaning. It was found through results that students were less anxious
and were more motivated with online classes.
The study claimed to improve the online experience by empowering students and
teachers with online technical types of equipment. The need to develop quality
content by faculties to facilitate online language learning skills was also expressed.
The researcher mentioned the need to conduct the study on a larger and more varied
population sample to get a better knowledge of the topic. [130]
76
Samrah Jamshaid et al. (2023), “Pre and Post-Pandemic (COVID-19) Mental Health
of International Students: Data from a Longitudinal Study”, studied the state of mind
of foreign students studying across the globe in various Universities in the early
phases of the pandemic as well as during the pandemic. Mental health was examined
concerning two parameters and those were anxiety and depression. A Longitudinal
method was applied. It was found that during the pre-pandemic stage female students
experienced the worst effects on mental health. There was an increase in the anxiety
and depression levels of students under study. In addition, the researcher specified
that there is a need to increase the number of counseling centers for students as the
pandemic has shown long-term effects on the mental conditions of the students. [131]
Reyhan Zeynep Pek et al. (2023), in “The Role of Machine-Learning in Identifying

Students at risk and Minimizing Failure”, used different machine-learning algorithms
to discover students who are prone to fail in exams. The results were compared and a
few better algorithms were selected. These algorithms were then used to develop a
blended model. At the front end, a web application was developed for collecting
responses from the students. The questions were based on the student's demographic
and academic information. Stratified K-fold validation and hyper parameter
optimization mechanisms were used. 94.8 % accuracy resulted when the both
academic and demographic dataset was used. Whereas with a single academic dataset,
an accuracy of 98.4% was achieved. [132]
HanjiyaLyu et al. (2023), in “Human behaviour in the time of COVID-19: Learning

from big data”, discussed the current big data mechanisms that are prominently used
to understand human behaviour during the period of the pandemic. Measure, model,
and influence were the three dimensions of big data, which were preferred. Text,
audio, and video data were analysed to measure behaviour. Further, the paper
summarized the existing models like compartmental, machine-learning, and hybrid
model. The challenges and opportunities of using these models were studied. [133]
Budi Tiara Novitasari et al. (2023), in “The influence of technology readiness and
learning motivation on students’ performance in online learning during the COVID 19
pandemic: A story of Indonesian Higher education students”, tried to find the
correlation between technology readiness amongst students and its effect on student
motivation and their performance. Data was collected from Four twenty-three
77
students using a survey method. The answers were analysed using partial least square-
Structural equation modelling. The result showed that the technology readiness of
students is an important factor affecting student motivation and performance. In
addition, it was proved that learner motivation influenced their performance. [134]
Bastian A. Betthauser et al. (2023), in “A systematic review and meta-analysis of

the evidence on learning during the COVID-19 pandemic”, surveyed school-going
children to find the influence of lockdown on the learning growth of students. The
study was conducted in 15 countries. The results showed that there was a 35%
shortfall in the learning progression for an academic year. The shortcomings were
particularly seen in students, with a low socio-economic background. The students
were found more deficient in understanding mathematics as compared to reading
skills. High-income, middle-income, and low-income countries were considered in the
study and it was noticed that the learning progress was low in middle-income
countries as compared to high-income countries. However, there was no prominent
evidence collected for low-income countries. [135]
Abdellah Ibrahim Mohammed Elfeky et al. (2023), in “Examining the effects of

virtual classroom use inside learning management systems on enhancing student
satisfaction”, experimented by integrating the learning management system with the
virtual classrooms. Two separate groups one of the boys and the other of the girls
were used in the study. The experimental results showed that both groups were
satisfied with the new integrated environment. In addition, the system proved better in
shaping the English language learning skills of the student groups. [136]
Rahim Sadigov et al. (2023), in “Deep learning based user experience evaluation in
distance learning”, the feelings of people were analysed through tweets posted on
social networking sites. The opinions were collected regarding the shift in the
education system. The proposed model reflected an accuracy of seventy-six percent. It
was noticed that around fifty-four percent of people expressed a negative attitude
towards the situation. The experimental results were compared for LSTM and
Recurrent neural networks. The LSTM model gained the best results. [137]
Alemayehu D. Taye et al.(2023), in “A machine-learning approach to predict self-

protecting behaviours during the early wave of the COVID-19 pandemic”, discovered
78
the factors that implied more responsible behaviour of people in protecting
themselves against the pandemic. A random forest algorithm was used in the study. It
was found that awareness about the disease; demography, socio-economic
background, age, and marital status were the key factors to decide the level of
prevention an individual took during the disaster. [138]
Shaun Liverpool et al. (2023), in “Mental health and wellbeing of further and higher
education students returning to face-to-face learning after COVID-19 restrictions”,
studied the mental state of the students as they returned to universities after the
pandemic for continuing education or for enrolling in higher education. Depression,
anxiety, and stress levels were counted for evaluating the mental health and well-
being factors. It was observed that the student's mental health concerns were reduced
as compared to the study undertaken during the pandemic.
Students at a post-graduate level were less mentally disturbed. It was suggested that
students coming from other states or regions for getting an education were more prone
to poor mental health conditions. Training students for making them self-efficient and
motivating them for improving their physical activities are some of the measures that
should be taken up at present for improving and maintaining the mental health of the
students. [139]
Yanyan Hu et al.(2023), in” Association between sleep quality and psychological

symptoms: A cross-sectional survey of Chinese university students performed during
the COVID-19 pandemic”, through this study it was found that there was a change in
the sleep pattern of students during the COVID pandemic compared to prior times.
The experiment was conducted on students with good, average, and bad amounts of
sleep. Less than two hours of screen time, and providing counselling to students with
low sleep quality were some of the suggestions given by the researcher to improve
sleep patterns. It was analysed that sleep quality directly affects the psychology of
students and interventions were needed to improve it. [140]
Xinchao Li et al. (2023), in “The effect of student’ online learning experience on

their satisfaction during the COVID-19 pandemic: The mediating role of preference”,
designed a hypothetical model to find students’ choice regarding online or offline
learning. Students took up the survey, and it was found through the response collected
that eighty percent of students preferred the offline mode whereas twenty percent
79
opted for the online mode. Internet access, cost of using online infrastructure, self-
confidence, teaching methods, involvement of students, and choice of instructions
were some of the attributes that contributed to finding the student's satisfaction with
their overall educational experience. [141]
Zi-Yu Huang et al. (2023), in “A study on computer vision for facial emotion
recognition”, designed a deep neural network to find out the most analytic features of
the human face. These could then be used to evaluate the emotions. The model was
realized using a convolutional neural network that was trained using the AffectNet
and Real-world Affective Faces Database. The observations made from the
experiment stated that the area around the nose and mouth contributed more while
forecasting facial emotions. It was concluded that the results of this work would help
improvise the precision achieved by computer vision models in the future. [142]
Sofia et al. (2023), in “ Machine-learning based model for detecting depression

during COVID-19 crisis”, conducted a survey containing twenty-one questions to find
the depression level among people because of the pandemic. The analysis of the
survey was made by using machine-learning algorithms like KNN, Decision Tree, and
Naive Bayes. The accuracy of the model using KNN was found to be more precise as
compared to other algorithms used in the study. The researcher suggested that there is
a need to replace the conventional methods of detecting depression with machine-
learning methodologies. The Hamilton rating scale measured the level of depression.
If the score was below seven no depression was analysed. While a score of 7-17 was
considered a person having mild depression, 18-24 as moderate, and 25 onwards was
treated as serious depression. [143]
Ignacio Aguaded et al. (2023), in “The impact of post-pandemic learning loss on

education development: A systematic review”, assessed the learning loss that students
are bearing after the pandemic. The assessment was done owing to different
parameters like technical resources, socio-economic status, inclusion criteria, the level
of connection between teachers and students, and loss due to lack of summer/
vacation programs. The research papers and articles were reviewed for each of these
categories. It was concluded by the researcher that the lack of communication
between students and teachers largely affected knowledge acquisition and
development among students. Further, it was observed that in some students this led
80
to symptoms of depression and anxiety. The digital competencies were lacking in
both teachers and students. In addition, there was a lack of a properly organized
program structure for online delivery of the content, as the academic world was not
prepared for this transit. [144]
Kardoyo et al. (2023), in “Hybrid, Online or Face-to-Face Learning? Experiences

during COVID-19 Pandemic”, captured the educational involvement of students. The
study was confined to the students of economics. It was revealed through the results
that they preferred offline classes. In addition, they feel that the technologies used
while imparting online or hybrid education should be continued while conducting
physical classroom lectures. The students felt that the teachers could do better
classroom management if they bring these technical skills into regular practice in the
classroom. [145]
Laia Subirats et al. (2023), in “Temporal analysis of academic performance in higher

education before, during and after COVID-19 confinement using artificial
intelligence”, experimented to find the learning outcomes of the children. The study
was done for the period of the pandemic as well as before and after the pandemic
phase. Three major groups of students were noticed.
The first category of students was those who studied regularly, and then the second
category belonged to those students who took studies seriously at the end of the
semester just one month before the exam. The third group consisted of children whose
performance did not improve throughout the semester. It was evident that students
studying regularly got good grades. Those doing last-minute preparation showed
inconsistent results. Hence, it was concluded that this category required interventions
so that manipulative practices would be brought under control. [146]
Lindita Bande et al. (2023), in “Lessons from Graduation Project during COVID-19
Pandemic for Future Applications, Methods, and Tools”, studied the effect of the
pandemic on the project development by engineering students, which is a part of their
four-year degree program. The online mode phase that emerged due to COVID-19,
then the constant shuffle between online- offline mode and finally resuming the
physical mode had affected student psychology to a wide extent, which was evident
from the results of the study. The research also evaluated the usage of various
technical tools in project development by the students during the pandemic phase. The
81
study tries to explore some methods that would turn out to be beneficial to students in
developing their graduation projects in a better manner irrespective of the mode of
education. [147]
Saud Altaf et al. (2023), in “A Hybrid Framework of Deep Learning Techniques to

Predict Online Performance of Learners during COVID-19 Pandemic”, generated a
framework to check the performance of students considering several attributes. The
model was developed using 1D Convolution Neural Network and LSTM. The
researcher claimed that the results of this model would be beneficial in generating a
hybrid model in the future, which could then be used to analyze student performance.
The proposed model reflects an accuracy of 98.8%. [148]
Nawaf Awadh Alreshidi et al. (2023), in “How to make Learning Process Smoother
for Students through Teaching Practices: Action Research”, taught the topics of
mathematics like ratios, proportions, and percentages. The experiment proved that
pre-requisite knowledge about the subject, proper planning for implementing the
syllabus, and teaching approach, helps the students to understand the subject concepts
easily. The research work has two parts. In the first half, the topics were taught to the
students with an unstructured approach, which resulted in poor performance, by the
students. After conducting the action, research by delivering the teaching in the
structured format it was found that the students' results improved. [149]
Nivin Hasan et al. (2023), “Evaluating the effectiveness of E-Learning in Private

Schools of Amman, Jordan During and After the Corona Pandemic”, discovered the
learning outcomes that resulted due to the e-learning method. It was found that though
physical lectures resumed still some best practices of the e-learning technology should
be integrated with the traditional teaching methods. This would improvise teacher
support. Further, it was recommended from the study that the students, teachers, and
parents should be educated about handling the e-learning tools so that the overall
education model could be effective. [150]
Mei Cao et al. (2023), in “The relationship between home-based physical activity and
general well-being among Chinese university students during the COVID-19
pandemic: the mediation effect of self-esteem”, carried out a survey to find if the
students’ overall welfare was dependent on their involvement in the physical routine.
82
The responses from the Chinese university students were collected. It was revealed
that those students who were regularly engaged in some or other physical activity
were able to acquire and maintain self- confidence in them.
Thus, it was proved that there was a direct correlation between physical activity and
well-being. In addition, the factor of self –worth played a vital role in deciding the
level of student wellbeing. It was suggested by the researcher that the students should
get encouragement from the family members about active participation in sports or
any activity that could be done at home. This would help them to gain confidence,
which would further help them to increase their overall development and stability.
[151]
Bin Gong et al. (2023), in “Machine-learning and data analysis-based study on the
health issues post-pandemic”, analyzed senior citizens, who were of the age of 50
above. It was seen in the pandemic there were people who were suffering from certain
health issues regardless of the pandemic. The researcher captured the impact of the
pandemic on those who were facing the health issues and the people who were not
having any serious ailment.
The lifestyle of the people was the deciding factor for finding the impact. As it was
found that the citizens who were engaged in daily physical schedule were able to
maintain stress levels and were mentally and physically fit. These people were also
experiencing good level of sleep. The use of machine-learning was made to frame a
hypothetical model. The researcher claimed that the use of such model would help to
understand the health parameters of the people from their medical history, and public
health databases.
It will help the health administrators to carry out effective procedures in case of
adversities. The responses of the people were collected in the four major categories
and those are physical, psychological, social, and environmental parameters. The
results were then interpreted by applying the statistical measures. It was concluded
that the reduced physical activities significantly triggered the stress levels and there
was a correlation between these two factors. [152]
2.3 Summary
The chapter gives the crux of the research carried out by researchers in the field
related to the topic of study. The literature review was helpful in understanding the
83
applications and working of several machine-learning algorithms. This helped the
researcher in finding the appropriate algorithms, which were then used in the research
work.
In addition, the comparative analysis of the algorithms could be understood from the
review. Further as per the demand of the research work, the deep learning algorithm
based papers were also reviewed. It helped in finding suitable facial expression
recognition algorithm that could be applied in the modular development. Several
survey based papers were analyzed so, as to understand the framing of the
questionnaires. Insights were gained about the psychometric scales, which could be
used in academics. Knowledge was gained about the python and other frameworks
that could be used for application development.
The impact of the pandemic on academics and the physical, motivational,
behavioral, co-curricular, and extracurricular parameters, which were the focus of
study, were reviewed separately. This was done to understand the amount of work
carried out by different researchers across the globe. It was observed that each of
these factors were analyzed separately or in combination of two or three parameters.
Thus, all the five factors, machine-learning, deep learning and the survey method are
merged and the researcher has attempted to provide it on a single platform.
84
Chapter 3
Machine-learning algorithms and technology for model formation
3.1 Introduction
Technology always updates at a faster rate as compared to any other field. The first
chapter described how education evolved and how slowly and unknowingly
technology became a part of the whole education system. Today, technology is a part
of our routine activities and we are on the path to experiencing the era of Data
Science. Data Science is not new, it came into existence around a half-century back
but now in the 21st century, it has become popular.
One of the important reasons for gaining this popularity is the generation of a huge
amount of data. A lot of internet access by people, and the use of social media
contributed to this data creation. Recently, in the COVID-19 pandemic when
everything shifted online there was a sudden increase in the usage of virtual platforms
in education as well as in other areas. Technology provided us with alternate solutions
for almost every activity that we were doing offline before the pandemic. Thus, data
started growing exponentially. This opened new opportunities in the field of data
science. The analysis and recognition of patterns from this data have become
important and could help in effective decision-making.
Artificial Intelligence and Machine-learning are playing dominant roles in many
domains. They are responsible for the growth and inventions in industries.
In the education industry, it is expected that AI-enabled tools will be regularly used.
Automating some of the monotonous tasks such as assignment checking, conduction
of unit assessments, student feedback, and evaluation of objective examinations could
be easily carried out with these tools. Because of this, the burden on the teachers
would be reduced to some extent and they can spend the time constructively to help
students acquire a better understanding of the subject concepts.
Many decisions are to be taken by the students during their education journey like
selecting a suitable career path. Consequently, they have to search for respective
courses to achieve the goal and identify their aptitude, attitude, strengths, and
drawbacks. Until the advent of ML and AI, we were making use of the computerized
psychometric test to identify the cognitive and other abilities of students. However,
nowadays, through AI-based tools, the personalization factor could be added.
85
Students can learn at their own pace as these tools adapt to the learning style and
frequency of the individual. These tools were available before the pandemic but
research showed that the use of these techniques accelerated owing to the pandemic. It
was observed that the online academic phase increased the acceptance level of
educators towards the technology and many researchers forecast the use of
technology-enabled tools in academics as a new normal. Teachers now rely on these
tools for virtual learning.
Surveys conducted by many researchers showed that educators are now ready to
imbibe these tools into the core educational framework. On the other hand, it is
important to measure how the student community is going through this transition.
Children of different ages, from varied economical and family backgrounds, faced the
pandemic situation and it would be fruitful to observe how they faced the educational
challenges during this phase of transformation.
The proposed model put forward by the researcher tried to capture the impact of the
pandemic on the physical, motivational, co-curricular, extracurricular and behavioral
patterns of the students. The model was constructed using machine-learning and deep
learning mechanisms. There are two modules in the model. The first module
computes the above factors using ML algorithms. While, the second module
simultaneously observes the facial expression of students. To do the analysis deep
learning algorithms were used. Convolution Neural Network (CNN) model was
developed.
3.2 The technology used in the Model formation

The model proposed by the researcher was developed using Machine-learning and
deep learning algorithms. Before we go into the detailed discussion on Machine-
learning (ML) and Deep Learning (DL), which is, briefed in this chapter it is
important to discuss AI, ML, and DL. There is a thin line of difference between these
techniques. Artificial intelligence came into existence in 1956.
AI symbolizes the big picture. Artificial Intelligence stimulates human behavior.
Human beings are versatile. We have the skill to interpret, respond, react and adjust
our behavior from experience and requirement. AI helps in developing computer
systems that can attain task, which requires human intelligence. Machine-learning is a
subset of AI. The Machine learns from previous data, improves from mistakes, and
makes predictions.ML generates fast and accurate outcomes.
86
Deep learning is a sub-part of Machine-learning. Deep learning provides algorithms
that are influenced by the structure and functionality of the human brain. Deep
learning shows better results when we have to deal with tremendous data. The time
required to train a deep learning model is more as compared to machine-learning
whereas the testing time for DL based model is less. The efficiency of deep learning
algorithms increases with a large amount of data. This is the reason that has made it
so popular in recent years.
Diagram of AI, ML, DL
Deep
Learning
Machine-
Learning
Artificial
Intelligence
Fig 3.1: Hierarchy of AI, ML and DL
3.3 The Role of AI in Education

In the years to come artificial intelligence will have a massive impact on education.
The field of education is no longer confined to learning management systems and
ICT. Leading-edge technology such as AI and ML modified educational tools and
transformed the overall framework in institutions and Universities. It is important to
make the students aware of the use of this technology, as shortly they will have to
work in an environment where AI will be the reality.
AI is not going to replace teachers but it is there to make the learning experience
better for children as well as for instructors. In the current scenario, we need a hybrid
model, which could combine the best features of the traditional model and artificial
intelligence-enabled systems. The objective of introducing these AI-based models is
to help teachers spend quality time with students where they can focus on the
individual needs of the students by giving them personalized attention.
Identifying the particular academic requirements of the individual child had been the
priority of educators for many years. AI will help in realizing this. Some companies
87
are developing platforms that enable AI-driven testing, evaluation, and feedback to
students. Some applications help students identify areas where they are lagging and
emphasize more learning on them. Thus, the learning progress of students would be
tracked and customized solutions could be provided. AI also gives access to Universal
learning, helping students learn through any language using the AI applications.
When students spend time in offline classes’ communication, collaboration, and
understanding through discussions should be focused on.
Applications of AI in Education Respective AI Tools used
Diagnostic assessment, analyzing individual Learning analytics,

needs and provide feedback adaptive learning methods
Assessments and grading AI grading applications
Computer vision, Remote

Conduction and Evaluation of exams
Proctor based systems
A/R, V/R, virtual labs, face

Smart classrooms and laboratories recognition, speech
recognition systems
Examine the past and present data of the student to

predict the future performance of the student or even Predictive Analysis
help predict the academic interest of the student.
Table 3.1: AI / ML Tools used in Education
3.4 Significance of Machine-learning in education

Today we are living in an era where, one is taking support of digital tools and
technology everywhere. With the advent of the pandemic, education has become more
diverse and sophisticated. Now, the software developers are able to offer solutions
that are more effective. These resources are now being used in improving classroom
instructions and student learning experience.
The working of the conventional computer algorithms is quite stereotypical. The
algorithms follow a structured approach. Machine-learning gains knowledge from
previously collected data. It examine the prior experiences and modifies the outcome
in the light of the most recent information. Thus, one can obtain the most pertinent
88
data from the program instead of the same results. This helps in improving website
experiences.
The pandemic imposed the need of introducing the ML approaches in education.
Before the pandemic there were ICT tools, smart classrooms, E-learning, distance
learning approaches, which were being used in education. However, they were not
utilized to a maximum extent. As, the pandemic struck then, all these methods were
used with maximum capacity. This was the period when everyone realized that
though the technology had evolved, we were not ready with a robust system, which
could manage education in adverse times.
After the pandemic a lot of research is going on to find ways in which we can imbibe
machine-learning in classroom teaching. This would not only make the fullest use of
existing ML techniques in education but will also make the education system more
profound.
Students are no longer passive listeners. They like interactive sessions, and have
questioning mind. Due to internet, they are well informed. Hence, the role teacher is
more challenging. Teachers must cater to the diverse demands of their students.
Especially after COVID, efforts are required to identify and make individualized
learning easier. The tools for education applications that use machine-learning can
accomplish this.
Teachers must cater to the diverse demands of their students. Tools for education
applications that use machine-learning can make individualized learning easier.
The software does real-time monitoring of student performance. It examines available
data to improve several facets of the educational process. It may also provide the
greatest resources for a research paper the student needs to write or propose the ideal
learning path for them. There is not enough time for teachers to work with every pupil
individually. Particularly in public schools with crammed classrooms, this is true. The
personalization of the educational process depends on machine-learning and software
built on this technology. [153]
3.5 AI and ML approaches in education
3.5.1 Learning Analytics

It is the process of collecting and analyzing data of students. It not only includes
academic information about the student but also gathers background information
89
regarding their hobbies and other aspects of their personality. Concerning academics,
one can analyze the learning styles and learning pace of the learner. Similarly, it helps
in understanding the subject interest of the student. These analytical tools help to
transform this data into an important knowledge base. This will lead to well-organized
and improved educational systems.
3.5.2 Objectives of Learning Analytics:

1. To assist students in acquiring skills and techniques that will help them in lifelong
learning.
2. Provides personalized feedback to students at regular intervals.
3. Supports the growth of certain abilities like teamwork, critical thinking,
communication, and creativity.
4. Track Student performance.
5. Encourages Self-awareness thereby helping students to work on themselves.
6. To make informed decisions related to syllabus structure.
7. To build an effective educational system that will promote innovations, high-
quality teaching, as well as learning.
3.5.3 Predictive Analytics:

Mining historical data to forecast the future is a part of predictive analysis. It takes
time to dissolve complex data. For this, it uses statistical methodologies and advanced
analytics. In terms of education, both faculty and teachers leave digital traces of data
at different points in the academic process.
Objectives of Predictive Analytics:

1. To determine the cause of absenteeism between students. It could be health or
financial concerns or a lack of understanding of the subject. It is important to
uncover the actual reason and provide assistance for it.
2. Every student has a particular learning deficiency, and different comprehension
and grasping capabilities. Predictive analysis helps to find children that require
special attention. There could be students who are doing very well in academics.
Teachers could push them to give their best. In contrast to this, there could be
some children displaying weak performance. Specialized courses could be set up
for these pupils. Specific data metrics can also be applied to locate their issues.
90
3. To pinpoint a sharp increase in the dropout rate. To assure consistent enrollment
in the particular department using predicted results.
4. To take into account the physical, sports, and extracurricular activities of students
from the previous database and pursue them to participate in inter-collegiate and
University-level competitions.
5. It helps to spot the most recent academic trends. It helps us to know certain things
like whether online classes are more effective than traditional learning or which
grading system is beneficial for students around.
3.5.4 Adaptive Learning

Adaptive learning platforms offer customized learning material to students based on
their learning abilities. The traditional notion of a single classroom and one learning
methodology for all students of one class is losing its appeal. In the ML approach,
efforts are made to attain a truly adaptive state. The ML learners continuously gather
the data in real-time and try to build analytical models that can give trustworthy
results. Adaptive Framework is classified into three major parts as explained below.
Fig. 3.2: Adaptive Logic Framework

Content:
91
Module structure
The modules are developed to help students understand complicated topics within a
subject. The concepts are subdivided to get a smaller problem that could be
understood by the students. Later the solution to smaller parts is synthesized to get a
solution to the complete problem. This method makes the student understand the
difficult concepts following a stepwise approach. This methodology allows the
student to apply knowledge gained to analogous situations.
Interpersonal skill development
This is a key measure for scaling the cognitive level and involvement of the student in
a particular course. Automated grouping of students with similar interests is also
facilitated by this system.
Content Compatibility
Adaptive systems should be able to handle both fixed and flexible content sequencing.
The selection of fixed or flexible sequencing depends on the assessment results.
Metadata
Many times while understanding some concepts we require an understanding of
related concepts. Metadata is defined as data that gives brief information about
interrelated concepts. Mostly, the adaptive systems use this kind of tagging
mechanism for making learning easier.
3.5.5 Assessment
Predictive psychometric pattern
A child could be accurately placed on individualized learning paths through truly
adaptive exams. Predictive values are incorporated by utilizing systems that
automatically choose questions based on replies from students. The system also
generates reports that link test results to a student’s instructional requirements.
Diagnostic classification
Models of diagnostic categorization assure whether a learner has mastered or not a
certain set of abilities.
Proximal development
The zone of proximal development ideally helps the instructor describe the gap
between what learners can accomplish on his or her own and what they can perform
without assistance.
Competency framework
92
The Competence of any student is the potential, skills or knowledge used to
demonstrate in real-life or applicative situations. The standard framework defined to
assess this competency in a neutral approach is called as Competency Framework.
The Competency Framework is built to assess person’s skills holistically, thus it not
only examines the technical skills but also reviews the softer skills. This assessment is
very useful to segregate students with higher competence who can have a higher
impact on overall society if given opportunities in the right domains.
3.6 Need of Artificial Intelligence and Machine-Learning

In today’s world, globally there is a buzz about the terms like Artificial Intelligence,
Machine-Learning and Data Science. With this buzz, also comes a lot of investment
in terms of efforts, money and time to explore and apply usage of these technologies
to solve the real-life challenges faced by the human life. It is believed that these
technologies would be able to solve complex challenges and problems, which the
conventional technologies may not be able to solve or take much longer time to solve.
We also observe many times that the terms Artificial Intelligence or most commonly
referred as AI and Machine-Learning, which is also referred as ML, are used
interchangeably. However, there is clear distinction between the two.
While, AI is a domain which focuses on mimicking the human intelligence through
machines that effectively tries to capture the vast and valuable experience gained by
human intelligence in the fields of Computation, Visualization, Perceptive Skills,
Design and Creative domains etc. through machines. Thus, ideally it tries to truly
replicate the comprehensive human intelligence with machines.
On the other side, ML is the subset of AI that works on machine-learning algorithms,
which can be trained using pre-existing relevant datasets or content. Thus, ML is a
discipline that processes huge amount of data to analyze and come up with
meaningful insights out of it; also in the process, it keeps learning itself and keep on
optimizing to get as closer to the reality as feasible.
While discussing about AI and ML, it would not be complete if there were no mention
of Data Science. Data Science is a very closely used term along with AI and ML. Data
Science is a field, which processes huge amount of data and analyzes to come out
with meaningful conclusions. While all these are interconnected topics, we will
largely focus on Machine-Learning in this research project.
93
In the above section, we briefly touched upon different technologies that help us
getting closer to mimic the human intelligence. Now, let us understand the real need
of using Machine-Learning answering why, how, where questions.
In the recent times especially during COVID19 pandemic and post that, almost all
businesses and educational institutes including government institutions have adopted
the digital platforms. This has given rise to generation of digital content exponentially
especially in last 3 years and it continues to grow with passing time even with higher
rate. According to one of the renowned reports, it has been predicted that globally the
digital adoption got a major boost between December 2019 to July 2020. During this
period, globally the digital adoption has increased from nearly 33% to around 56%,
thus we have seen a rise by almost 70%.
It has given the world an opportunity to efficiently capitalize this huge data; process it
and analyze to derive valuable conclusions out of this exercise. Technologies
especially, Machine-Learning are playing a crucial role in achieving this objective,
given that we use an appropriate model in an effective manner.
Each ML algorithm can be considered to be a pipeline having defined with set of
programming operations, which when run would result into an ML model. This model
generated from a learning process, is based on mathematical & statistical rules &
algorithms and tries to represent the real world scenario. We can consider this as a
pipeline which receives data from one side, it is capitalized, processed and analyzed
using the in-built libraries & algorithms and provides the desired outcomes in the
forms of clusters, classifications, sentiments and recommendations. There are
different models used for different purposes.
Some of such popular models used in the industry are Supervised Learning,
Unsupervised Learning and Reinforcement Learning models etc. The suitability of the
performance of each model can be assessed based on certain assessment metrics,
which are Precision, Recall and Accuracy. Machine-Learning is a very competent
tool, which is used to provide resolution to the real world challenges. The success of
effective resolution of such challenge depends on identifying the right challenge or
scenario to use Machine-Learning technology and model.
Typically, when we deal with the real-life, industrial challenges, they appear to be too
complex to get an appropriate solution. While, the past experience working with ML
highlights us that it is a very effective tool to manage such complex topics, which may
be too difficult for the conventional software applications or tools. While the
94
conventional tools dealing with relatively simpler problems may provide precise
solutions, ML could be very effective in solving complex challenges due it its ability
to provide probabilistic solutions, which is found to be the right approach to chase
such challenges. Some such examples include, solving communication problems
(detected while using some communication tools) using say a Chabot application.
Deep Learning model of Machine-Learning appears to be the most suitable tool for
such communicative issues having large number of representations and dimensions.
Another type of problem that can best be addressed by ML, are multi-parameters
problems. One more example that can be considered suitable for leveraging the
Machine-Learning model is a visual challenge. For example, if we need to predict the
behavior of the person based on his or her facial expressions, the Supervised Model in
which learning through the preexisting datasets of people emotions vs the facial
expressions can be captured and the results could be effectively predicted.
Some more complex examples of the real-life situation are – predicting the prices of
real estate or properties in various parts of the city or predicting the share prices inside
the stock market.
3.6.1 Benefits to the users:

Once we apply the ML models in our software applications, it can provide following
benefits to the users:
Classification: Based on the inputs the user has provided, it will classify the input
data into predefined classes e.g. 0 or 1, true or false, spam or non-spam etc.
Selection of Top Items: Based on the search criteria provided by the user, arranging
the order of the search result in such a way that the most relevant results are ordered
at the top of the list for the convenience of the user
Automation: Certain repetitive scenarios or a set of instructions or steps performed
by the user are automated. E.g. based on user’s daily routine of referring to maps to
go to a workplace finding the most optimum path that helps reaching the workplace
faster, providing an automatic alert with this optimum path at an the right time.
Fraud and Anomaly Detection: There has been an exponential growth in digital
transactions more recently especially post COVID times. This has also given rise to
increased number of fraudulent or suspicious financial transactions. ML can help to
categorize each transaction into valid or fraudulent / suspicious transaction and raising
alert to the user at the right time in case the transaction is detected to be fraudulent.
95
Recommendation: While the user is searching a specific product in a catalog having
huge collection, suggesting the user intended product based on the product description
and user’s earlier search criteria.
Creation of content: Generating content based on user inputs and learning from
earlier use-cases, e.g. if a user desires a specific image in the context of some location
and specific product, then generating a suitable user intended image.
Extraction of Information and Tagging: identifying the most important information
from texts or images, tagging that information, and providing notification to relevant
users. The information could include the names of included people, location etc.
Responding to the Questions: Which provides relevant responses to the queries
posed like “Does the video or audio queried by the user, provides relevant information
about the subject that it is referring to”.
In the table below, we can see the features or functionalities provided by Machine-
Learning technology:
Fig 3.3: Features of AI / ML
3.6.2 Benefits to the Businesses and Institutions:

Following are the benefits that can be reaped using Machine-Learning tool by the
businesses or sectors mentioned below:
96
Finance: It helps with better financial projection and analyzing potential risks
associated with the targeted market and industry. Based on the financial projection
and risk analysis, businesses can take calculated decisions in choosing the right
segments for investment to get higher level of returns on their investment and to
safeguard the company assets. It can also predict what is trending in current times for
better results.
Healthcare: Based on the patients past medical history, current symptoms and
individual genetics, it can predict the possibilities of critical illnesses like cancer,
heart complications in advance and provide guidance to eliminate or reduce the risk of
developing it in future. It can also suggest personalized healthcare solutions and
medications for the treatment. Apart from this, it can be paired with the health apps to
track patient’s health conditions on daily basis and providing alert in case some health
parameters appear to be abnormal.
Transportation: Based on the desired route for the traveler, it can analyze the
optimum path to avoid heavy traffic areas and reach the destination in least feasible
time. This can prove to be quite beneficial for the transportation business in terms of
saving fuel required for the travel and in turn the money.
Marketing: Based on the market research and consumer purchasing style and
preferences, it can predict the right consumer base for the product. This will help the
Marketing firms to design the marketing campaigns targeting the specific consumer
base learning from their past purchasing history. It can also help to find the recent
trends and preferences.
Web Page Designing: Machine-Learning can help to identify the maximum
appealing content of the web pages, which is often visited by different set of users.
Based on this analysis, the graphic design and content of the pages can be enhanced to
attract more users connecting to the website. In turn will help the ranking of the
website in different search engines.
Data Mining: Machine-Learning can be used to deduce vital information from huge
content. Using this information, the data scientists can suggest interested and newer
consumer base, find trending patterns and help the businesses to enhance their
operations.
Employee Relations: For the HR department, it can predict the star performers in the
organization as well as the domains or departments where there could be higher level
of attrition based on the skillset of the employees working in those departments and
97
the pull from the market for the skills. It can then suggest the learning modules for the
existing employees for upskilling in case of loss of talent and suggest the suitable
industry segments from where the employees with desired skills can be hired.
Logistics: Learning from the experience, analysis of entire supply chain can help to
predict optimum level of inventory.
Education: Education can take benefit of Machine-Learning in multi-fold levels. It
can help students to identify the upcoming skills and domains that are highly desirable
from the industries, which in turn will help them to take right decision in choosing
their careers. The learning can help the institutions to identify bright students as well
as students needing support. This will help the institution to device varied
methodologies and ways to personalize learning content and processes for the
students.
3.6.3 Significance of using Machine-Learning in current research project:

The current research topic is in the domain of Education and we have seen in the
above-mentioned section, how Machine-Learning can add value to this sector. This
research project aims to analyze the positive and negative impacts of COVID on the
students thereby analyzing their educational preferences. For that, the researcher had
decided to review the facial expressions of the students derived from the videos. The
researcher also had chosen to use the survey for getting the inputs from varied
categories of students to understand the impacts on them.
To work on this research project, there were two choices to select from – one was the
conventional way and the other one was to use the modern and trending approach of
using Machine-Learning. Within conventional approach too, there were options like
complete manual approach or a hybrid approach where some of the manual efforts
could have been saved by conventional technologies. Had we considered any of the
traditional ways for analyzing these inputs to derive the results and to recommend the
actions for improving the gaps in the online learning approach, then it would have
taken huge amount of time and effort; also, it would not have been much effective,
precise and efficient.
Thus, the researcher had chosen to apply the Machine-Learning approach in this
research project. Apart from the benefits that are mentioned above for applying this
approach, in the process there has been a lot of learning for the researcher.
98
3.7 Machine-Learning assistance to teachers
Students are no longer passive listeners. They like interactive sessions, and have
questioning mind. Due to internet, they are well informed. Hence, the role teacher is
more challenging. Teachers must cater to the diverse demands of their students.
Especially after COVID, efforts are required to identify and make individualized
learning easier. The tools available for education applications that use machine-
learning can accomplish this.
It has been observed that post pandemic the students have varied requirements. Thus,
the time has come where the personalization of the educational process is to be made.
This is not possible within classrooms where the student strength is more. The
teachers do not get enough time to work with every individual. Thus, with the help of
machine-learning, software could be developed and we can come up with a better
educational framework.
3.8 Machine-learning in Behavioral Analysis:

In classroom, teaching the teachers can understand the behavior of students to some
extent through the interactions they have with the students and general observations in
the classroom. In addition, some students participate in certain activities so, the
teachers can understand their interactive, and co-operative skills. However, there are
limitations to this. The teachers are not able to interact with everyone in the class.
With changing times, students started using online platforms for learning. Owing to
this, many data were generated. Machine-learning algorithms are required to be
trained first. For this, we require data. With education being transited to online mode,
an opportunity was created for implementing machine-learning algorithms in
behavioral analysis.
According to theories of behavioral science, the students use learning platforms at
their own pace. They make use of the available learning modules as per their
individual needs. This became evident at the time of the pandemic when the students
actually used online mode of education. If we see to this with a positive perspective,
researchers now have the opportunity to analyze the data generated and make use of
the machine-learning models for predicting student behavior. This predictive analysis
will definitely help the teaching fraternity to adapt a better approach for teaching the
generations to come.
Following are the ways to monitor online behavior:
99
1. Path or Pattern of accessing online resources
This is one the basic behavior that can be analyzed by counting the number of times
the student logs in to a particular resource. Similarly, logging out of the resource can
be counted. Counting the number of times the students toggle between the pages, or
modules is also noted.
2. Social interaction
The participants in online learning environment are dependent on network. They can
communicate and interact in a collaborative way. The students can post their doubts in
the doubt forum, which can then be answered by the teachers or even other fellow
students. In contrast to this, in an offline environment, students can approach the
teachers. If they are unsure about a topic, they can consult the instructor
independently.
The prominent factor in case of online mode is the attitude of the student in getting
his/her problem solved. This mostly depends upon the individual personality.
It has been observed that in online social media behaviors, while submitting text-
based content for learning, students typically reflect their own knowledge in a self-
organized manner. To some extent, the learner characteristics could be observed
through their social media posts and interactions. The areas that interest them could
also be found through social platforms.
3. Resource Utilization
The content that the children are accessing online plays a vital role to interpret the
temperament of the online learner. Apart from this, the resources, which they are
using, are also important. There are videos, text, research papers, online newspapers,
magazines, journals, blogs and varied resources, which are available.
While watching a video content one has the option of replay, pause, rewind or fast-
forward. In the textual data, we can have page up, page down or preview options.
4. Assessment and Response
In digital environment, the evaluation was based on objective exam pattern. The
drawback of the objective examination is that it makes the students focus on the
result-oriented approach. They develop a tendency to ignore the process or concepts
of the subject.
The online assessments store the final marks or grades of the learners. In contrast to
the traditional method where the children write, a subjective answers. At end of the
100
exam, we have the answer sheets of all the subjects. The handwriting of the students,
the way of detailing of the answers also helps in knowing their aptitude.
The manner in which the students responded to the online assessment was also a
major factor that should be considered for gaining insights about the behavior of the
student.
3.9 Machine-learning in Motivational Analysis

The next parameter that the researcher studied was the level of student motivation.
According to the literature, motivation and engagement are relative terms, but there is
a thin difference between these terms. A student can be motivated but it does not
mean that he/she is engaged. To engage the students, it is necessary to make them
participate in some subject related activities.
Motivation is an energy that can be felt by a person. Primarily motivation can be a
trait or it can be a state. A “trait” refers to a habit that has become a part of the
person’s personality. A “state” is a temporary phase. Thus, it not last for long.
Two basic types of motivation are explained below
1. Intrinsic Motivation
This type of motivation depends upon natural factors. Curiosity, interest in a
particular area, excitement about learning something is some of the factors that keep
people motivated about a task or work or education. The satisfaction and confidence
that one gets while performing or completing a task is also an element of motivation
for some people. Research tells that the intrinsic motivation decreases as time passes.
The curiosity of a child decreases as he/she grows. One of the reasons of losing
intrinsic motivation is irrelativeness between what they learn in schools and what is
practiced in daily life. The stress by teachers and on performance and academic
grades is also one of the factors that decrease intrinsic motivation of the children to
some extent.
Above all students will remain naturally motivated only if they learn the things for the
sake of learning and not for displaying something. Sometimes too much importance is
given to external factors, which lead to loss of interest of students in studies.
2. Extrinsic Motivation
Rewards, awards, monetary gain, praise, privileges are some of the factors that
describe external motivation. Studies reveal that extrinsic motivational factors
contribute and help teachers and parents to motivate the students. However, the work
101
done with extrinsic motivation cannot generate confidence or a sense of
accomplishment among students.
3. Feedback Mechanisms for measuring motivation
Feedback helps us to analyze the student’s understanding, satisfaction, interest and
several other factors, which can be used to evaluate the motivation level of the
students. Motivation could be further classified based on levels of motivation as low,
intermediate and high.
4. Approaches of collecting Feedback
Two primary ways in which feedback could be collected are the closed ended
approaches and textual approaches.
5. Closed Ended Method:
In this type, MCQ or objective questions are asked to the respondents. The researcher
designs the questionnaire. Hence, the researcher has the control on the type of
questions to be framed. More or less number of questions could be framed depending
on the requirement Specific category to be focused can be decided and accordingly
the items in the survey could be added. In addition to spoken and textual feedback,
closed-ended techniques can be employed with any form of input, including video or
graphical analysis.
The Likert scale is the most popular mechanism under closed ended approach to get
feedback.
Text analysis is a cutting-edge technique for collecting feedback. In the machine-
learning approach, a model is trained on text that is similar to the feedback. The
model is then used to forecast a result. The classifier that is generated by a machine
teaching model trains, tests, and cross-validates itself. The model can then be applied
on live textual feedback and results could be predicted. Text-based techniques can be
used by training a machine-learning model to forecast an outcome from a set of texts.
The machine-learning model designed in the study analyzes the student responses and
identifies the level of motivation of the student in relation to the effect that the
COVID-pandemic has imposed on them. Accordingly, Positive, Negative, or Neutral
effect of pandemic on student motivation has been estimated through the study.
3.10 Physical Analysis

The Human brain has trillions of junctions that manages millions of functions. There
are specific hormones that are released when one does physical activity. Physical
102
activities help maintain hormonal balance in the body and contributes in coordinating
the physical and mental health of the person. There are three basic and natural traits
found in human brain.
Firstly, the brain experiences pain if they found someone in pain. Thus, the emotion
of empathy is generated. Secondly, the brain starts thinking and focusing on things or
ideas, which one imagines or visualizes. It works towards realizing it. Hence, it is
important to think positive. As the way we think manifests into reality. Thirdly, the
brain generates physical pain from an emotional hurt. It is important to have mental
balance, so that physical health is maintained. It is important to have compassion,
gratitude, forgiveness and healing abilities. These capabilities makes people mentally
and emotionally strong, which is reflected, in physical wellbeing. Research have
shown that there is correlation between mental and physical health.
To analyze the effect on the physical parameter during the pandemic, the researcher
considered the amount of time children spent on exercise or any other physical routine
like yoga or home-based physical activity. The second factor that was noted was the
amount of sleep that the child could get. The eating habits and the number of times
the children were consuming meals was also recorded. The factor whether, the student
was infected with COVID and experienced some physical changes was also
considered. The responses of all these parameters were fed to the machine-learning
algorithms and then the results were analyzed to find the effect on the physical
parameter.
3.11 Analyzing the co-curricular and extracurricular factors

The pandemic had slowed down the physical activities of the students. Throughout
the year, students are involved in some or other co-curricular or extracurricular
activities in the schools/colleges. Co-curricular activity participation is mandatory to
some extent as the students get credit points for it. These credits are counted in their
exam grades. Extracurricular activity involvement is by choice of the learner. Some
students pursue sports or any other art form and participate at inter-collegiate,
University, zonal, district, state, National or international level. All these activities
help in maintaining their physical fitness. Above all, it builds confidence, and
containment and mental stability in them. Mental stability again leads to good
physical health.
103
All these factors are interrelated to some extent. Their interdependency is discussed
below
3.12 The interdependency between the five parameters and the overall effect
In the above sections, we have observed the impact of online learning on the students
on five different parameters i.e. Physical, Motivational, Behavioral, Emotional, Co-
curricular / Extracurricular independently. However, as these parameters are exhibited
simultaneously by anyone, the collated effect of online learning on these parameters
would be important to study. Machine-Learning models would help in this analysis.
3.13 Working of Machine-Learning:

The Computer gives output as it is programmed to generate it for specific input. The
logic is important for any computer program and the algorithm provides the basic
steps that govern the program. When it comes to Machine-learning the computer is
provided with data (input) and examples of output. As a result, we get a
program/Model. These models can be used for prediction, decision-making, and
problem solving. Through ML algorithms, we might want to increase the accuracy of
prediction or improve the efficiency of problem solving. Depending on what we want
to achieve the performance measure is defined.
Fig. 3.4: Schematic Representation of Machine-learning System

The above diagram depicts the basics of the machine-learning system. The learner
accepts the data, considers the previous knowledge database, and the model is
designed. The reasoning works with the model and given a new problem or task
comes out with a solution.
104
Steps to create learner in the Machine-Learning System:
1. Collect the data for training.
2. Find the objective function.
3. Select the appropriate representation for the function.
4. Search for a proper learning algorithm.
The data that we gather for training has various features. We must take into account
relevant features depending on what we want the model to interpret. Another
important task is to find a proper function that could be applied to the dataset.
3.14 Broad Categories of Machine-learning:
3.14.1 Supervised Learning

In this method, the learning takes place in a guided and controlled environment. The
machine is fed with labeled data. In this type of learning the data comprise input along
with corresponding output. For every input x, the machine is fed with corresponding
output y. From this, the machine designs a model so that given a new observation x
the machine interprets its corresponding output. For example, if the machine is given
four types of flowers hibiscus, lotus, rose, and lily, as input then the machine is
explicitly trained to understand the features or attributes of an individual flower. The
machine is provided with labeled output so the machine could map the features of
each flower (X values) and would be able to put it under the proper label.
Important terminologies
1. Features
This is an important term that we need to understand to implement machine-learning
algorithms. Usually, the machine-learning algorithm consists of a training phase and a
testing phase. The data that we give as input to the machine can be considered as an
instance. The values that describe this instance were termed as features, attributes, or
characteristics. It is important to express these feature values. The feature could be
Categorical- Blood group (“A”,” B”, ”AB”, “O”), Gender (“Male”, “Female”).
Ordinal- It is a descriptive feature such as “Agree”, “Strongly Agree”, or “Disagree”.
Integer-valued- The example for this is the number of words in the text.
Real-valued- Height is an example of a real-valued feature.
2. Training and Testing Phase
105
The dataset used for training is the set of examples. A single example called an
instance could be described with the help of features of that instance and a
corresponding output feature is provided during this phase. The Complete dataset is
divided into two parts “Training” and “Testing”. From the attributes, we have to set a
target attribute. Suppose we consider the attributes as “X” and the target attribute as
“X1”. In training, we feed both “X” and “X1” to the model. Thus, the model
understands that for a particular set of features what outcome is generated.
In this phase, only input values for a particular instance are provided and the target
feature is to be predicted. Suppose “Y” represents the testing domain and “Y1” is the
target attribute to be predicted. Here “Y1” will not be fed to the machine. The
variance between the predicted target attribute and the actual target attribute will be
monitored. This observation decides the accuracy of the designed model. Then
depending on the output data value, we decide whether it is a classification or
regression problem.
Fig. 3.5: Machine-Learning Phases
Overfitting and Under-fitting:

The output of the training phase is a model or function. If we consider an X-Y plot in
which we get a graphical representation that tries to cover all data points then it is
termed as overfitting. Overfitting indicates that the Model was trained considering a
large number of features and precise features. This leads to wrong predictions. In
contrast to this when the model is trained using a very less number of features we
106
come across the problem of under-fitting. Under-fitting represents a line in an X-Y
plot where the data points hardly lie on the line.
Supervised learning approaches
Depending on the output data value, we decide whether it is a classification or
regression problem.
Classification
The classification algorithm is preferred when the output (Y value) is discrete. For
example given the symptoms of the patient (X value) predict the disease (Y value) of
the patient. Some of the popular classification algorithms are K-Nearest Neighbor
(KNN), Support vector machine (SVM), Logistic Regression, and Naïve Bayes. These
algorithms are widely used in pattern recognition, sentiment analysis, text analysis,
and email filtering to detect if the mail is spam or non-spam. They are also used in
face or object recognition.
Regression
Regression algorithms are helpful for continuous values that occur in the output. If the
input (X) is given as the various features like location, floor area, number of
bedrooms, etc., and related to this we want the machine to predict the price of the
house (Y value) then in this example as the price is a continuous value regression
algorithm will be used. Linear Regression, Support vector regression and Lasso
regression are a few types of regression algorithms.
3.14.2 Unsupervised Learning

This type of learning is not directed or guided. The machine has to understand the
dataset and find out the hidden patterns to recognize the output. The data provided to
the machines is unlabeled. The output is not specified. The machine is only fed with
an input value that is “x-value” the machine has to interpret the corresponding “y-
value” from the experience. The features of the input data points are observed and
objects with similar properties (features) could be clustered.
Unsupervised learning approaches
Clustering
The process of grouping objects or data points having similar features is called
clustering. The need to cluster data is to find the underlying patterns in the data. It
also helps to identify the attributes of different clusters. Clustering is beneficial to find
customer segmentation, arranging books on similar topics, and many such sorting-
107
related activities. While forming clusters it is important to define the similarity
measure. For example, if a teacher asks the students to arrange blocks of similar
shapes together then it is vital to state the similarity measure. In this example, the
measure of similarity could be size, shape, color, or all three attributes.
Association
This is a type of unsupervised learning where the algorithms try to explore the
associations between the data. We consider an example of a supermarket where we
find that customers purchase bread, butter, milk, and jam or others purchase bread,
butter, fruits, and milk. In the above scenario, the algorithm tries to take into account
the maximum associated data in this case we can find that bread, butter, and milk are
strongly associated. Thus if someone buys bread the further suggestion from the
machine could be for purchasing butter and milk. In the case of an offline store, the
vendor could arrange bread, butter, and milk on the same or nearby shelves, which
prompts the customer to purchase these items together. This is a type of market-basket
analysis that is performed. Apriori and FP-Growth algorithms are some of the types
of association algorithms.
3.14.3 Reinforcement Learning

This learning strategy is helpful in dynamic and uncertain environments. It is used to
solve complex problems. The major elements of reinforcement learning are the agent
(learner), the environment (with which the agent interacts) the policy that the agent
follows to take action, and finally one needs to observe the consequences after taking
action. The agent can get a reward or penalty. The learner continues to interact with
the environment to understand given a state what action could be taken rather it tries
to get an optimized solution.
3.15 Data Pre-processing for the proposed model

The machine-learning algorithms work well on preprocessed data. It is an important
step for getting accurate results. Primarily, two datasets were used for training the
model. The Kaggle dataset was used for drawing the physical and motivation-related
attributes. The second database was used for filtering the features corresponding to
the behavior. These two datasets were then merged and a new dataset named
concatenated dataset was created. This data was then divided into the 80:20 ratio.
Eighty percent of the dataset was used for training and twenty percent for testing.
108
Fig. 3.6: Data Preprocessing of the Model
Feature Extraction from the concatenated dataset

The data comprised of questions in the form of attributes. These questions were
designed for capturing the students' responses to their physical activities such as
eating, sleeping habits, and amount of exercise. Some questions were used for
capturing their behavior, social contact, and motivation resources. The age of the
student is the prime attribute. A list of conditions is framed wherein the category of
109
student is decided that depends on the age. Four conditions were considered. First less
than twelve years, second less than or equal to thirteen and greater than or equal to
fifteen, third greater or equal to sixteen and less than or equal to eighteen, fourth
greater or equal to nineteen. Corresponding to these conditions a list of values was
assigned. This contains “ES” –early school, “JHS” – Junior high school, “SHS” –
Senior high school, and “College”. A new column was added in which we store the
category of the student as per the conditions.
Data Pre Processing for capturing Physical Impact
From the concatenated data, five attributes were observed for physical impact. Firstly,
the column stores response related to the sleep of the student. The attempt had been
made to map the age of the student and the time he/she was spending on sleep. A new
column “Student has normal sleep” was designed where the condition and value were
passed and it was observed if the child had normal sleep or not. The default value was
set to “-1”. For all the four conditions related to age, the value assigned is “0” which
was interpreted as normal sleep. For the second attribute that counts on hours spent on
physical fitness, the conditions were again mapped with the age. The values assigned
were “1”, “2”, “3”, and “4” and the default value set was “0”. The new column
“Fitness awareness” was passed with the conditions and values. The third attribute
stores the count for the number of meals per day during the lockdown. The new
column “Eating habits” was formed for this. The values assigned are “0”, and “-1”.
The default value assigned was “1”. The fourth attribute examines the weight change.
Here encoding was done by replacing weight gain with “1”, “-1” for decreased
weight, and “0” if the weight remained constant. The fifth column evaluates if the
student came across any health issues. “0” replaced the response “Yes” was replaced
by “1” and “No”. The values of all the above five columns were summed up and
passed to the final input data frame. Conditions were framed for finding the physical
changes and were assigned to final input. If the five attributes is less than one then a
negative impact was considered. For sum greater than or equal to one and less than
two, neutral effect was evaluated and the sum greater than or equal to two, positive
effects were interpreted. The values set were “-1”,”0”, and “1” for negative, neutral,
and positive effects respectively.
Pre-Processing data for finding the effect on Motivation
Few attributes were monitored from the merged dataset for estimating motivational
changes. The first column evaluated the social connectivity of students with family,
110
friends, and relatives. “1” and-1 respectively replaced the responses “Yes”, and “No”.
Time spent on social media and time spent watching television were combined. A
new column for storing the time spent on entertainment examined the number of
hours utilized for television and social media. The condition was passed to the final
input data frame for examining if the number of hours the student spent on
entertainment was more or less than time for self-study. A new column was created to
capture the total study time of the student. For this time spent on online classes and
time spent on self-study, these two attributes were analyzed and summed up. The
next attribute collects data in hours about how well the time was utilized. The last
column stores the rating of the online class experience. The responses were in the
string format, which was labeled in numeric terms. Very poor, poor, average, good,
and excellent were replaced by “1”, “2”, “3”, “4”, and “5”. The addition of all the
above column values was taken. Final input data frame was passed with conditions
and values. For motivational changes < 3 the value assigned was “-1” which reflected
a negative effect. Similarly, for 3 <=motivational changes < 6, “0” was allocated that
corresponded to a neutral effect, and for motivational changes >6 positive effects
were reflected and the value “1” was used to represent it.
Feature extraction for behavioral Analysis
For behavioral analysis, labels replaced the responses. The labels were as follows;
strongly disagree - ‘2’, disagree -‘1’, uncertain-‘0’, agree, - ‘-1’, strongly agree- ‘-2’.
Exposure to electronic screens, social personality, laziness due to staying at home,
Prolonged use of E-learning tools leading to nervousness, buying of digital tools
unaffordable to students, not in favor of online learning, lockdown measures leading
to frustration, assignment submission online was stressful, Face-to-face interaction
contributes to boosting students’ academic achievements, taking online exams and
quizzes was not comfortable. The responses to these attributes were added.
3.16 Machine-learning algorithms implemented in the proposed model

In all four models are developed using four different algorithms. KNN, SVM, RF and
NN algorithms were used to develop the models. The prediction efficiency of these
four models was then analyzed and the best model was implemented.
The generalized workflow of the four models (KNN, SVM, RF, and NN)
The data used for the model creation is data.csv and COVID-19 Survey Student
Responses.csv
111
1. In the first step data processing is done. The features are extracted from the
dataset and are stored in features_ex.py. Using this file and some required inputs
data is generated.
2. The features are extracted from the data_path.
3. The command x, y_p, y_m, y_b= extract_features(data_path) is used.
4. The data was splitted into training and testing datasets.
5. x_train, x_test, y_train, y_test=train_test_split(x, y_b, test_size=0.2,
random_state=42), here y_b is for behavior.
6. x_train, x_test, y_train, y_test = train_test_split(x, y_m, test_size=0.2,
7. random_state=42), here y_m is for motivation.
8. x_train, x_test, y_train, y_test = train_test_split(x, y_p, test_size=0.2,
random_state=42), here y_p is for physical.
9. Similar procedure was applied for physical and motivational models.
10. Then, by using the command model=train_model(data_path, model_path) models
were trained and saved. It was followed for all three categories.
The train_test_split function

A dataset is divided into two subsets using this function in the
sklearn.model_selection module. One is used for training and one for testing. A
machine-learning model is trained using a training set, and its performance is assessed
using a testing set. In the research study, the data of the dataset was split into a ratio of
80:20. Eighty percent data was used for training and the remaining twenty percent
was utilized for testing.
The steps 6, 7, 8 described above explain how the data was divided as x_train, and
x_test. The datasets were divided, as X and Y, and the amount of the dataset to be
allotted to the testing set is test_size. For all the models, the test_size attribute is set
as 0.2. This indicates that 20 percent data was used for testing and the training was
done using the remaining data.
One can train and test the model using the subsets of the data after it has been divided.
However, each time when the code is executed, the results can be different. To avoid
this the “random_state" parameter is used. [154]
Random State
It is used to reproduce the same set of train and test data points every time the data is
split. If we don’t use this attribute then in every run some different we will get
112
different train and test data splits. This will not be helpful, as debugging would
become difficult. If no random state is provided the system will use a random state
that is generated internally. So, when we run the code multiple times different train
and test dataset set will be reflected. Thus, the model ends up reflecting unpredictable
behavior.
Thus, “random_state should be assigned with an integer number. It is required

attribute. It helps in reproducing and debugging the code. Comparison of the results
and performance evaluation of the models is also possible because of the
“random_state” attribute. [154]
3.16.1 K-Nearest Neighbor (KNN) Algorithm

The algorithm is a type of supervised learning algorithm mostly used to solve a
classification problem. It attempts to classify the data by observing how its nearby
data has been classified. KNN falls into the category of the lazy algorithm. It does not
learn a discriminative function from the training dataset. All the machine-learning
algorithms are divided into training and testing phases. KNN algorithm does not train
the model in the training phase rather it stores the examples and later at the time of
testing identifies the example from the training dataset that is closer to the test data for
which the prediction is to be made. The algorithm is suitable for a small database. The
data should be labeled and noise-free. It depends on feature similarity so one can do
classification using the KNN classifier. The value of K is to be predefined and it
should not be too small or large as it affects the outcome. Moreover, the value of k
should not be biased in one direction. If the k value is very small then the bias is too
noisy. If it is very large then we have to face processing and resource issues. The
value of K is selected in proportion to the total number of database records. The
process of choosing the appropriate value of K is defined as “Parameter Tuning” and
it is important for better prediction accuracy. To choose the correct value for k we use
the formula sqrt (n), where n= Total number of data points. Generally, odd value of K
is chosen to prevent confusion between two classes of data.
113
Fig 3.7: Implementing KNN Classifier in the model
Implementing KNN Classifier in the model

To implement the KNN algorithm we require setting certain parameters. In the model,
the researcher had set the ‘n_neighbors’ value as 10 for estimating the physical and
motivational changes. To understand the behavioral variations 20 nearest neighbors
are studied. It is crucial to fix the value of K. The weight parameter is used in
prediction. The possible values of the weight parameter are “uniform” and “distance”.
Uniform weights consider all points in each neighborhood to be weighted equally. In
the proposed model, the value of weight is equal to “distance”. Due to this, the nearby
data points that have more influence on a new data point will be taken into account.
The distance can be Manhattan or Euclidean distance. Generally, Euclidean distance
is used on real values otherwise hamming distance is used on binary or categorical
data. The research framework uses the Euclidean distance method to classify the new
value into the appropriate class.
Once the data is preprocessed, a classifier is constructed. For the classifier new
function, “classify” will be created. This function takes the items, new items to be
classified, and the value of “k” as input. Existing closest neighbors are stored in the
list called neighbors. Each element holds two values. One for the distance from the
new data item and the other value to represent the class of the neighbor. The distance
is computed by the generalized Euclidean formula. The k nearest neighbors and their
114
class is examined. The class that appears the maximum number of times is selected
and that will be the predicted class for the new item.
Euclidean Distance formula
Process of Updating Neighbors

Add a new entry to the list of “neighbors” with a specific distance. This list should
have a maximum length of k. if it is less we add the item to the list even if it is far
away. The list should be filled with k before rejecting the item. To add a new item to
the list, check if the item’s distance is less than the item with the maximum distance
in the list. The new value will take place of the item with the largest distance. The list
is sorted in ascending order so that the item with the greatest distance can be found
easily.
Determining the Neighbor Class
The class that appears most frequently in the “neighbors” should be evaluated. A
function count is used with class names that exist in neighbors serving as the keys. A
key will be added if it is absent, otherwise its value will be incremented.
FindMax ( )
The value of count ( ) is passed to the FindMax ( ) function and then max value is
returned by this function.
Generalized steps to implement KNN:
1. The value of K is a positive integer.
2. Select the K- records in the database that are closer to the new data point.
3. The algorithm predicts the output of the new data point by examining the output
of the K nearest data point.
4. The new data point is classified according to the most common classification of
the K nearest values.
Advantages of KNN
1. No need to train
2. It effectively works on and manages large datasets.
3. It is accurate and effective.
4. It is non-parametric
5. It is simple and easy to understand.
Drawbacks of KNN:
115
1. The computation of KNN algorithm is costly.
2. The selection of “k” should be precise.
3. It is dependent on Euclidian distance.
4. It is quite sensitive to the data, which is prominently different from the rest of the
data.
5. If the number of data points in every class varies then the KNN algorithm results
are affected. [156]
The next model was developed using the Random forest algorithm. This algorithm is
discussed below.
3.16.2 Random Forest:

This algorithm is suitable for both classification and regression problems and a
statistically driven algorithm. It is based on the concept of decision tree algorithms.
Decision tree algorithms are easy to build and interpret but they are less accurate on
real data. The decision tree tends to be useful in the training dataset. It cannot give
efficient results in classifying new data. This is the reason that we move on to
Random Forest. The algorithm has two random processes. First bootstrapping and
second is feature selection. So, the word “random” was significantly used. It makes
use of multiple decision trees hence the word “forest” is used. Thus, the algorithm is
named as “Random Forest”. [158]
Difference between Decision trees and Random Forest

Decision tress are highly sensitive to the training data. This could result in high
variance. On the other hand, random forest is the collection of multiple random
decision trees. It is less sensitive to the training dataset. If the decision tree is allowed
to expand in an uncontrolled manner, it typically experiences overfitting problem.
Overfitting is prevented by using random forests, which are built from subsets of data
and produce results based on average or majority rating.
Concept of Ensemble Learning:
In real life, we see that there could be different approaches to solving a single
problem. It depends on the People’s perspective toward that problem. Similarly when
someone has to take a certain decision one has to consider all the factors that
contribute to making the decision turn beneficial to the decision-maker. On the other
hand, one has to think over all the negative aspects and be ready with some alternative
116
if ever it has to be faced after taking the decision. Thus, usually people try to consult
those who have already faced similar situations and try to examine their approach to
solving the issue. Ensemble learning works on different datasets or subsets of
datasets. There could be hundreds of algorithms working under one Random Forest.
Ensemble learning combines the results from various individual models and predicts
the result. [159]
Bagging and Bootstrapping:
Bootstrapping: The process of creating subsets from the original dataset is termed
bootstrapping.
In the process of bagging, the data is bootstrapped and its aggregate is used to make
decisions. Random subsets are generated from the original dataset. Suppose we have
K-learning algorithms we could split the original dataset into K subsets. The problem
arises if the dataset is small then the partitioned subset that will be provided as a
training dataset for the learner algorithm will be small thus the model created will not
be able to predict the generalized results properly. Thus, we cannot have an
independent subset of the dataset. However, we can sample randomly from the data
wherein there could be some data in common. The subset could have redundant
instances, as well as some instances, that do not appear in the particular subset, which
leads to different subsets. All the subsets will be of the same size. We apply random
sampling with replacement to obtain a dataset with different values.
117
Fig. 3.8: Decision Tree from Bootstrapped Dataset
Creation of Decision tree from Bootstrapped Dataset:
For creating a decision, tree random columns are selected from the bootstrapped
dataset. On the next node, the columns that were considered at the root will be hidden
and some other columns will be taken to split the data. Likewise, different decision
trees are built. For identifying the best columns, one can apply algorithms such as
GNA index, variance reduction, or entropy. Because of this most of the decision, trees
will vary. Multiple decision trees are created.
Significance of Bootstrapping and feature selection
Bootstrapping ensures that we are not using the same data for every tree. It helps the
model to be less sensitive to the original training data. The random feature selection
helps to reduce the correlation between the trees. If we use all the features or same
features then most of the decision tress will have the same decision nodes and will act
in a similar manner. One more aspect is some of the trees will be trained on less
relevant features so they will give predictions accordingly. However, there could be
trees, which give bad prediction. Therefore, the overall effect is balanced.
118
Counting votes for Prediction:
The result predicted by each decision tree will be checked and counted. These
predictions are called votes. For example, if we want the decision tree to predict if the
person (new data) has been diagnosed with diabetes or not. We will apply every
decision tree on this instance for which the result is to be predicted. If we get 95 votes
out of 100, which predict that the person has diabetes, and 5 votes that say that the
person does not have diabetes then the result or class depends on the maximum votes.
Voting is the measure that gives high accuracy when we use Random Forest.
Fig. 3.9: Decision Tree from Random Forest
Applying Random Forest in the study:

To execute the Random forest algorithm the data path is acquired from the file. The
next step is to provide a path for ‘extract_feature’ file.
Steps to Create Random Forest
1. Build new dataset from the original data
2. Randomly select rows to form the new dataset.
3. Every dataset contains the same number of rows as the original dataset.
4. Random sampling with replacement is applied
5. After the 4th step we get bootstrapped data.
6. Train the decision tree on each of the bootstrapped dataset.
7. Randomly generate a subset of features for each tree and then train the trees.
8. The tress are then built using the data and the feature subsets.
9. The results from each tree are evaluated and combined.
119
10. As it is a classification problem, majority voting is considered and the result is
predicted.
This process of combining results is called as aggregation. Thus, we first do
bootstrapping and then aggregation. Together, the process is called bagging.
Parameters used in the Random Forest Model
In the random forest classifier, the hyper parameters are utilized to either speed up the
model or improve its performance and predictive ability.
Following are the parameters used in the model:
1. n_estimators:
It corresponds to the number of trees which will be generated by the algorithm. For
the Random forest model used in the study the n_estimator value is set to 200.
2. Criterion:
It helps to decide how to split the node in each tree. The supported options for
criterion are “entropy”, “gini”, “log_loss”. By default it is set to “gini”. In the model
it is assigned as “entropy”.
Advantages of Random Forest
1. It lessens the problem of overfitting in decision tree and increases accuracy.
2. It is useful and adaptable to solve problems involving classification and regression.
3. Both categorical and continuous values can be used with it
4. It automates filling in data's missing values.
5. Data normalization is not necessary because a rule-based methodology is used.
Limitations of Random Forest
1. It uses a lot of resources and computer effort to build multiple trees and integrate
their outcomes.
2. Training takes a lot of time, as it integrates numerous decision trees to decide the
class.
It also suffers from interpretability issues and is unable to establish the relative
importance of each variable because of the ensemble of decision trees. [160]
3.16.3 Neural Network:

Neural networks are capable of solving highly complex problems. These networks are
inspired by the human brain. The human brain contains several neurons of the order
of tens of billions. The neurons are interconnected to form a network. Individually,
the neurons can perform simple computations. Together they work in parallel to
120
achieve a complex task. They have distributed associative memory. They are efficient
in handling multi-dimensional data.
Structure of neuron:
Fig. 3.10: Neural Network Working
Nucleus: The neuron cell body consists of the nucleus.

Dendrites: These are used to accept the input.
Axon: Output is generated through this tail. The output is then forwarded to another
neuron through the synaptic layer.
An artificial neural network in which the basic units are input, node, and output
simulates the neuron structure. The computation of the node first computes the
weighted sum of the inputs and then applies a transformation function. Following this,
the output is generated.
Layer: The neuron is an input point and the layer stores these input points before
passing to the next layer.
Perceptron: The perceptron plays the role of a neuron in an artificial neural network.
The perceptron takes n inputs. Perceptron unit has two parts. In the first part, the
weighted summation of the input is computed. There is another unit called bias. (Phi
explanation).The transfer function could be a linear function just the summation will
be passed.
The layers of nodes that connect the activation functions (neurons in the brain) to one
another. The output of one node is an input of another since nodes are connected to
one another. A node receives weighted inputs, which are subsequently added together
and to which the activation function is then applied. The result will then be the
weighted sum of the inputs supplied to the activation function. [161]
121
The Artificial Neural Network structure has three subtypes
1. Single layer perceptron
2. Multi layer perceptron
3. Recurrent network
The perceptron theory was given by the American Psychologist Frank Rosenblat.
Single layer Perceptron
The layer is formed by taking processing element and combining it with other
processing elements. The oulput layer neurons have the processing capability. Thus,
the input is collected at the ouput layer and multiplied with the weights and
summation is done. Next the bias, activation function will be applied at the output
layer. Input is just forwarded to the output layer. We use this type of network when
the problem is linearly seperable. That is if one has to solve a classification problem
then single layer network could be used. In this network the input and output are
linked together. Inputs are connected to the processing nodes with various weights
resulting a series of output one per node. There no hidden layers in this type of
network. It is not efficient in all the areas.
Multi layer Perceptron
It is formed by interconnecting several layers. Multiple layers exists between input
and output layers. These are known as hidden layers. Each neuron of the hidden layer
can have different or same activation function. It depends upon the problem domain.
Input layers receives input and buffers input signal. The output layer generates the
output. There could no hidden layer or several hidden layers in a network. If there are
many hidden layers , the network complexity increases. But complex networks are
capable to generate efficient output.
Recurrent Neural Network
A sort of artificial neural network called a recurrent neural network (RNN) is used to
process data sequences. They perform particularly well for tasks requiring sequences,
including voice, time series data, natural language processing, and other operations.
This type of network saves the output of a particular layer and then sends it back to
the input layer, in order to get the output of the layer.
Working of RNN
The neural network's input is received by a layer, Suppose we consider this layer as
'x'. This layer processes the input it before sending it to laye the middle layer.
122
The middle layer is made of Multiple hidden layers, each with its own activation
functions, weights, and biases. Here if represent the middle layer by 'h'. Recurrent
neural networks can be used with neural networks without memory, meaning that the
various parameters of different hidden layers are not influenced by the previous layer.
So that each hidden layer has the same characteristics, the recurrent neural network
will standardize the various activation functions, weights, and biases. Then, it will
build one hidden layer and loop over it as many times as necessary rather than several
hidden layers. [155]
MLP Classifier parameters
Activation Function: It issued for normalizing the computed input to produce an
output. There are various activation functions such as linear function, Sigmoid,
Softmax.
These are the functions available in sklearn for activating the hidden layer.
1.”identity” , it returns f(x)=x. it is used to create a linear bottleneck. It is an no-op
activation function.
2. The logistic sigmoid function, "logistic," yields f(x) = 1 / (1 + exp(-x)).
3. The hyperbolic tan function, "tanh," produces the result f(x) = tanh(x).
4. The rectified linear unit function, or "relu," produces the result f(x) = max(0, x).
In the model “identity” activity function was used. The default activation function is
“relu”.
the weight optimization solver.
Solver
This parameter is used for weight optimization
1. An optimizer in the family of quasi-Newton methods is called "lbfgs." The
performance of “lbfgs” is better over other optimizers. The convergence
capability is also quick.
2. Stochastic gradient descent is referred to as "sgd."
3. “Adam” optimizer was used in the model. Kingma, Diederik, and Jimmy Ba's
stochastic gradient-based optimizer, is called as "adam”. It is a default optimizer.
It is efficient on large datasets. Especially for a dataset with around thousand
records or more. It is convienet to use considering the time required for model
training as well the validation score.
Learning Rate:
Following are the learning rate which could be used for weight updates.
123
1. Constant: The 'learning_rate_init' provides a constant learning rate called
'constant'. It is the default option available.
2. Invscaling: The 'invscaling' uses an inverse scaling exponent of 'power_t' to
gradually reduce the learning rate at each time step 't'. Effective learning rate equals
learning rate initial divided by power (t, power_t).
3. Adaptive: The 'Adaptive' maintains the learning rate at 'learning_rate_init' so long
as the training loss declines. The current learning rate is divided by 5 whenever two
successive epochs fail to reduce training loss by at least tol (tolerance for the
optimization) or, in the case of "early_stopping," fail to increase validation score by at
least tol. In the model “adaptive” learning rate was used.
Random_state
Determines train-test splitting if early stopping is utilized, random number generation
for weight and bias initialization, and batch sampling when solver='sgd' or 'adam'. If
one wants consistent results across several function calls, pass an int.
max_iter
It is set to an integer number, the default value is 200. It specifies the maximum
number of cycles. The solver iterates upto the number iterations assigned to it or until
convergence (specified bt ‘tol’). The number of iterations corresponds to the number
of epochs for the stochastic solvers (‘sgd’, ‘adam’), not the number of gradient steps.
Advantages of Neural network
1. Neural networks are quite flexible. They can adjust their parameters with respect
to the new data. It helps in
Adaptability – They are highly adaptable as their parameters are adjusted as per the
response thus they keep learning and improving.
Nonlinearity – They can model non-linear connection across input data. Thus, they
are suitable for applications, which may be able to work with linear models.
Parallel processing – They can perform multiple complex computations
simultaneously making them most suitable for applications like autonomous cars,
speech recognition etc.
Fault tolerance – They are not so susceptible to the faults within the input data and
thus quite robust. Thus generally they still produce fair result even if some input data
is corrupted or missing. [157]
3.16.4 Support Vector Machine:
124
This supervised learning algorithm can solve the classification and Regression
problem statements. Suppose we want to classify strawberries and apples. First, we
input the data for training the model. Whenever new data is fed into the model, it
should be able to classify the shape properly. Generally, a decision boundary or
hyperplane is drawn that provides a partition between the classes.
In this example, there are two classes and a hyperplane is drawn. For each class, the
data point, which is closer to the opponent class, is identified and a line is drawn close
to these extreme data points in both classes. These lines are parallel to the hyperplane.
They generate distances out of which one is positive and the other is negative. “D+” is
the shortest distance, which is closest to the positive point, and “D-“is the shortest
distance closest to the negative point.
These positive and negative distances are summed up and the resultant distance is
termed as margin. Margin is an important terminology and is required in deciding on
an optimal hyperplane for the classification to take place correctly. If the hyperplane
is not identified properly or we select a hyperplane with a lower margin then there is a
high probability of misclassification. The extreme data points, which were considered
while drawing lines parallel to the hyperplane, are called support vectors. [162]
Linear SVM:
When the data points get segregated into two classes just by drawing a hyperplane is
defined as linearly separable data. Linear SVM is applied to such data.
Non-Linear SVM:
This algorithm can be applied to non-linear data. The basic principle is to convert data
in low dimensional vector space to high dimension using the Kernel function.
Figure to be drawn before linear SVM:
125
Fig. 3.11: Non Linear SVM
There are various parameters available in Support vector classifier. In the model we
have used the regularization parameter (C), and the Gamma parameter. The parameter
options for SVC are given below:
1. The Regularization parameter which is denoted by “C’
2. The Gamma parameter
3. Kernel
4. Degree
5. Random state
The Regularization parameter
The fundamental use of the ‘C’ parameter in Support vector classifier, is to handle the
penalty parameter of the error term. It can be regarded as the level of classification
accuracy that the algorithm must achieve or the level of SVM optimization that it
must achieve. It regulates the mechanism of accurately classifying training points and
a rounded decision boundary. It simply, advises the model to use data points as a
126
support vector. If the value of C is chosen high, the model will choose more data
points as a support vector. This will give a bigger variance and lower bias, which
could result in the overfitting issue. If the model chooses lower value of ‘C’ then the
model identifies few data points as support vectors and thus gets lower variance and
high bias. Thus, the value of ‘C’ should be optimal to get better results.
The Gamma parameter
1. The Gaussian RBF (Radial basis function) Kernel employs the gamma parameter.
2. If the linear or polynomial kernel is selected, then we require a ‘C’
hyperparameter and no gamma.
3. It decides the degree of curvature for the decision border.
4. It gives more curvature with a high gamma value
5. It gives less curvature due to a low gamma value.
Advantages of SVM:
1. It is effective in high dimensional instances.
2. It can be used on semi-structured and unstructured data (images, text, or trees).
3. A subset of training points is selected from the decision function. These are called
as support vectors.
4. It allows using custom kernels. Thus, varied kernel functions can be specified for
the decision functions.
Disadvantages of SVM:
1. It takes long to train the SVM classifier on large datasets.
2. It is sensitive to outliers.
3. It is mandatory to perform feature scaling of variables before applying SVM.
It take time and effort to finely tune the hyperparameters “C” and “gamma”.
3.17 Summary
This chapter briefly explained the recent and trending technologies like Artificial
Intelligence (AI), Machine-Learning (ML), Deep Learning (DL) and Data Science
and their significance in education. It has discussed different machine-learning
approaches like Supervised Learning, Unsupervised Learning and Reinforcement
Learning; models like K-Nearest Neighbor (KNN), Neural Network (NN), Random
Forest (RF) and Support Vector Machine (SVM). The advantages and limitations of
each of the algorithm are given in the chapter. The data pre-processing for the model
and the implementation procedure of all the four models has been written. Each of the
127
five parameters taken for the study are described in detail. The effect of pandemic on
each of these factors and its evaluation through machine-learning is briefed here. In
addition, the benefits of using Machine-Learning for the businesses / institutions as
well as the end users have been provided. One of the major objectives of this chapter
was to study the significance of Machine-Learning application to the Education sector
and thus check the feasibility of its application to the research work under discussion.
All these factors are also described in details in this chapter to conclude that Machine-
Learning is the optimal choice to proceed ahead with the research work.
128
CHAPTER 4
FACIAL EXPRESSION RECOGNITION MODEL
4.1 Introduction
In Chapter 3, Machine-Learning algorithms were applied to monitor the changes in
the physical habits, behavioral influences, motivation, and participation in co-
curricular and extracurricular activities of the students. The second sub-module of the
research study focuses on capturing the facial expressions of the user while they are
answering the questions related to the above parameters. This chapter briefs about the
model formation for facial expression recognition and thereby evaluating the
emotions of the student.
Among all the features, the facial expression is the most important aspect to
understand the feedback of the students about classroom teaching. In addition, this is
the major factor helping the teachers understand the attitude, aptitude, emotional,
logical, and several other factors of a students’ personality.
The prior research carried in child psychology and the work done in pedagogy reflects
largely that there exist a co-relation between the personality of the student and their
academic performance as well as their overall success.
The research objective is to develop a student survey application. Through this, the
researcher proposes to collect responses of the students. While, the students fill the
survey, facial expression module captures their live expressions. This is an attempt to
add genuineness to the responses that the students are filling.
Another aspect of introducing this module in the study is the shift to online education
during COVID. This tool tries to capture the influence of the pandemic on the
education of the students. It analyzes the five major factors. In addition, the researcher
tried to find the extent of influence the pandemic has made on the students’.
There are several methods of face detection and face recognition. In this chapter, the
different approaches to face recognition have been explained. Further, the chapter
explains the algorithms and classifiers, which were used in the face expression
recognition module.
The Deep learning approach is used in this model formation. This had been briefed in
the chapter.
4.2 Computer Vision and Image Processing
129
Image Processing and Computer vision are often used interchangeably. Both of these
perform some kind of computation on images. There is a difference between these
two processes. Image processing works on the images. It tries to make changes in the
input image. Various processes like smoothing, increasing the brightness of the
image, highlighting the edges of the image, adjusting the contrast, and blurring the
image could be done. The output of this is always an image.
Figure 4.1: Image Processing – Blurred Image to Processed Image

Computer vision, on the other hand, tries to explore what the machine could see in the
image. It focuses on object identification from the image. In addition, it searches for
categories, labels, and coordinates from the input. Thus, the effort is made towards
getting some knowledge from the input image. The processing done in computer
vision is task specific. Computer vision systems rarely use raw images rather they
depend on image processing algorithms.
Similarly, it has been observed that image processing also uses computer vision to
improvise images. For example, face beautification filters, smoothing filters, and
bilateral filters are chosen in a particular manner and applied to the image. Improving
eye clarity, detecting facial landmarks, or tuning the image locally are some of the
advanced applications of computer vision. In most cases, computer vision and image
processing are used together. Machine-learning is one of the key branches of
computer vision. It is true that computer vision heavily depends on machine-learning.
For example, a model could tell the difference between a cat and a dog once it has
been trained rigorously to do so with those images. Many advanced image-processing
methods also use machine-learning models to convert images by applying filters like
artistic filters. There are certain filters used for refining the image quality. Often we
require studying the details of the image to enhance the performance of the computer
vision task. There is no hard line between these two approaches, which could
130
differentiate them completely. The difference reduces further when we do pixel-by-
pixel transformations. If the model is generating per-pixel analysis of a given input
image then it could be termed image processing.
Another view to do this is, such kind of computation needs image understanding.
Trying to find what that image is about. This reflects the use of computer vision.
Another example that could explain the interplay of these two methods is the use of
Convolution neural networks (CNNs). CNNs generally work on pixel intensity values
and try to process them to accomplish some task like the image or object recognition
from the input.
Internal layers are added to the CNN model, which would be considered as an image
filter with adaptive parameters. Thus, the work done by CNN can be considered as
some sort of image processing. The applications of CNN are not limited to processing
images. They show outstanding results in audio, video, and even natural language
processing applications.
Figure 4.2: Computer Vision and Machine-Learning
4.3 Emotion Recognition

The ability to be able to express what one feels through emotions is the most
fundamental skill in humans. In addition, the understanding of another person’s
emotions plays a crucial role in any organization as well as in our daily life. Being
131
sensible to emotions often helps in gaining valuable feedbacks about an individual,
organization or a system as a whole. Behavior is one of the important factors that
could be scaled and measured based on the emotional analysis.
In education industry, feedback from students plays a vital role. There are several
aspects like syllabus restructuring, measuring the grasping capacity of the students
related to different subjects, subject knowledge application, practical skills which are
dependent on the attitude, behavior or temperament of the student. After all the
approach of an individual towards doing, a certain task definitely matters while
counting the success or failure of that activity.
According to many researchers, it has been proved and believed that the emotional
state of the student during their learning period definitely influences the overall result
either in a positive or negative way. We have seen that in the pandemic phase some
students faced hard time and had to take a lot of effort to manage resources and
acquire basic knowledge. [166]
4.4 Teacher’s perspective in understanding emotions of the student

It is important for the teacher to know if the students are able to understand the topic.
Human faces work as a window of expressions. Through the body language, posture,
facial expressions one could definitely read out if the child is interested in the class. In
addition, the attentiveness of the student could also be noticed. This was very easy in
the offline mode. Teachers could change their teaching approach or could add more
examples if they felt that majority of the class was not getting the content.
However, in the online mode students started doing things mechanistically. Copy,
paste options became easy for them. If any question was asked, they could easily
google and tell the answer. On the other hand, there were students who were not at all
interested in class participation. While some were hesitant.
Slowly, it was recognized that we were just providing the content and information to
the students online. However, if really learning was taking place was a big question.
In addition, the factor of physical observations of students by teachers was missing in
the online method.
4.5 Different methods used in past few decades to understand emotions

Blooms taxonomy tells us about the three basic learning categories. These are
cognitive, affective, and psychomotor domains. In the recent years, with the advent of
132
Artificial Intelligence a lot of research is carried out to understand the relation
between the cognitive and affective fields. In addition, adaptive systems could be
wisely developed to control the affective element. Facial expression recognition,
physiological signal recognition, speech recognition and text analysis are some of the
major ways to detect the emotions. [174] [175]
Physiology deals with emotion detection by monitoring various biological parameters
of the human body. Electroencephalography (EEG), Photoplethysmography (PPG),
Galvanic Skin Response (GSR).Heart rate and pulse rate recording, blood pressure
measurement, eye movement, and overall body movement analysis. Nervous system
impulses could also be measured by observing the changes in those impulses.
Wearable devices are available which are used for collecting the physiological
signals. There are machine-learning algorithms, which can be implied for doing the
face, speech and text analysis. [173]
One of the prime dimensions of the research work is to understand the changes in the
behavioral patterns of the student and understand the temperament of the student.
Attempt has been made to gain understanding of this parameter. Facial expression
analysis was considered as the predominant factor that contributed and supported the
behavioral study.
The survey designed by the researcher is aimed to collect the student responses
regarding their online learning experience. Answering the questionnaire is one part
but in addition to this, the experiment takes help of continuous video streaming to
capture their emotions while filling the survey. Thus, the effort has been made to
capture the true real-time feelings of the students while they were responding to the
questions. Facial emotion recognition is one of the several techniques for recognizing
and analyzing emotions.
4.6 Facial Expression Methods

We can analyze the facial expressions through two basic methods.
4.7 Facial action coding System

In this method, the facial muscle movement is examined to recognize the emotions. It
is used in psychology to understand the expressions. The method consists of 44 action
units that were designed and grouped based on the human facial observations.
4.8 Facial emotion Recognition Structure
133
The emotion recognition system is a three-step process that consists of Face detection,
feature extraction, and then emotion classification. To detect the emotions of the
individual from the video it is important to recognize the face of that person. Some
fundamental steps are to be followed for recognizing the face as given below:
4.8.1 Face detection

This task could be performed due to the existence of computer vision. In this, we get
to know if a face is present in the image, video, or photo. Further one can also identify
the location of the face in the image. The technique of face identification is applied in
various fields like attendance, entertainment, security, healthcare and the list goes on.
Autofocus in digital cameras, crowd analysis are some of the popular application of
face detection. Apart from this there are challenges that are to be faced by the
algorithms such as occlusion detection, face makeup or face accessories, face
orientation, light effects on the image etc.
Figure 4.3: Face Detection Technique [146]
Timeline / Evolution of face detection algorithms

1. Viola Jones/ Haar-cascade algorithm (2001)
It used the concept of integral images. It was fast and accurate in calculations. This
algorithm was the first to gain commercial acceptance. It did not work well with
images where the face was covered, or if the face was not with a frontal view. It is the
most efficient algorithm. The algorithm is capable to identify the faces from real-time
video streaming.
It is the most popular technique used in face detection. It was unable to recognize
faces when they were not in in normal position. Like it was difficult to identify a face
that was turned or if it was upside down or in any other position. In addition, if the
person in image was wearing mask or glasses it was difficult to detect the face.
2. DlibHOG (2005)
134
It utilizes the histogram of oriented gradients or HOG features. The method is to
divide an image into grids, apply gradients on parts, and then create histograms.
Further, a linear classifier is applied on the features to understand if it is a “Face” or
“No Face”. It is a robust algorithm then Haar-cascade as it uses five filters like, front
looking, left looking, right looking, and front looking with right rotation, front looking
with left rotation. It analyzes every pixel within an image, to find the nearby
comparatively darker pixels. It then assigns an arrow towards the flow of darkness in
relation to the pixel. Every pixel undergoes this process of assigning an oriented
gradient to the pixel, by examining the pixel around it.
The HOG (I) function accepts an input of the image “I”; it will replace each pixel with
an arrow. Thus, gradients=Arrows. The gradients depict how light and dark move in
and out of an image. Complex features, such as eyes, may produce excessive
gradients. Thus, one has to aggregate the entire HOG (I) to create a “global
representation”. The image is divided into 16x16 squares. Each square is given an
aggregate gradient (G’), where G” is a function that could be max (), min (), or some
other function.
Although the algorithm works well in many applications, it still makes use of hand-
coded features that suffer from background noise and distractions.
3. SSD
The emergence of deep learning introduced algorithms that are more robust like
Single Shot face detector.
4. R-CNN
Selective search is used by R-CNN to produce bounding boxes, or regions. This
method evaluates the image via windows of various sizes and attempts to identify
objects by grouping nearby pixels together according to texture, intensity at each size,
and color. It creates a collection of areas for bounding boxes. To determine what type
of object an image in a bounding box is, it is passed through a pre-trained neural
network and then an SVM. Once the object has been identified, run the box through a
linear regression model to produce more precise coordinates for the box. [163]
4.8.2 Feature Segmentation / Extraction

Sometimes it can be difficult to identify the face and requires clarity in certain
features. Face Recognition is a task that goes a step ahead to understand whose face is
135
being detected. Hence, the head position, face light, gender, age and several such
attributes are considered.
4.8.3 Face Recognition

It is a tedious task. Lot of research is done to improvise the face recognition
techniques. New methodologies are implemented to get more accurate procedures that
would help in better face recognition. [165]
 Face Recognition techniques

Recognizing face and the research related to the analysis of facial expression,
facial emotion and similar topics have gained popularity in recent years. Researchers
from the field of computer vision, pattern recognition, psychology, neural networks
and computer graphics find this topic interesting and important for their respective
domains.
Face recognition is classified into four major categories as mentioned below
 Holistic Matching
 Feature based
 Model based
 Hybrid Methods
 Holistic Matching
The complete face is considered as input and is fed into the system. There are
certain methods, which are used in this approach, which are explained below: [164]
1. Eigen Faces:
This method deals with the two-dimensional problem of face recognition. It
constructs an Eigen face by inserting a set of photos into the database. This group of
images is called the training set since it will be used to compare and construct Eigen
faces from the images.
By removing distinguishing characteristics from the faces, Eigen faces are created. To
align the eyes and mouths, the input photos are normalized. After that, they are
resized to be the same size. In order to extract Eigen faces from the image data, a
mathematical process is used. This process is called as PCA.
A vector of weights will now be used to represent each image. The system is now
prepared to take requests. Finding the weight of the incoming unknown image and
comparing it to the weights of the existing images.
136
The closest-weighted image in the database will be shown to the user as a hit. [164]
2. Principal Component Analysis (PCA):
It is a dimensionality reduction technique used in probability and statistics. It is
commonly used in data science and machine-learning applications. PCA is a very well
established strategy. It has a source way back in 1901 in Pearson paper. The ML and
DL approaches have simplified the face recognition. Eigen face method using PCA is
an old method to identify faces.
Any matrix can be subjected to principal component analysis (PCA), which produces
a set of vectors known as the principal components. Each principal component is the
same length as the matrix's columns. The vector dot product of any two primary
components from the same matrix is zero since they are orthogonal to one another. To
represent each column of the matrix as a linear combination (i.e., weighted sum) of
the principal components, the various principal components created a vector space.
[164]
 Feature Based
In this method the position, the look of local features like nose, mouth, and eyes
are extracted before being sent to the structural classifier. When the system seeks to
retrieve features that are unseen because of significant alterations, such as head Pose
while matching a frontal image with a profile image, feature "restoration" poses a
difficulty for feature extraction methods.
Different method to extract features are used. Firstly, a common method is dependent
on edges, curves, and lines. Feature-template based method and structural matching
are the other two methods used in this technique.
 Model-based method
It is an approach based on models. The method aims to model a face. The
parameters of the model are utilized to recognize the image when the new sample has
been introduced. The model-based methods can be divided into 2D or 3D categories.
 Hybrid Method
This combines holistic and feature extraction techniques. Typically, these
technologies use 3D images. A face's image is captured in 3D to highlight the
contours of the eye sockets, chin, and forehead. Because the system uses depth and an
axis of measurement, which provides it enough information to generate a full face,
137
even a profile of a face would be sufficient. The detection, position, measurement,
representation, and matching components of the 3D system.
Detection: The act of capturing a face by scanning a snapshot or taking a live photo of
someone's face.
Position: Determining the head's position, size, and angle.
Measurement: Assigning measures to each facial contour in order to create a template
is called measuring.
Representation: It entails transforming the template
4.9 Neural Network

The human nervous system consists of the central nervous system, the spinal cord,
and the peripheral nervous system. The neuron is an important component of this
system and exists majorly in the central nervous system. There are sensory neurons
situated in the sensory organs. A neural network is a circuit of neurons. Today, we
come across an Artificial Neural Network (ANN), which is built using artificial
neurons or nodes. Thus, a neural network could be either a biological neural network
or an artificial neural network. The connections between biological neurons are
modeled in an artificial network as weights between nodes. The biological neural
network has synapses that pass signals to the neurons in the next layer. Neurons are
connected in various patterns. In addition, the output of the set of neurons becomes
the input of the other layer. The excitatory synapse is triggered and activates other
neurons. This mechanism is termed an excitatory connection. On to this, there are
inhibitory synapses that are less active and reduce the chances of passing information
onto the next layer of neurons.
 Importance of CNN in image recognition
Multi-layer perceptrons are not good at image processing, as a lot of computation
is required. Many attributes are involved even for a small image. To employ MLP it is
necessary to flatten the image. This reduces spatial information and these further
results in poor accuracy. The detailed information within the image is lost. Because of
these limitations, MLP is not widely used in the field of image recognition. CNNs are
efficient enough to preserve the spatial information of the image, at least when the
pictures are captured and fed in the original format. CNNs also use less number of
parameters compared to MLP.
 Difference between grayscale and RGB images
138
An image is made up of pixels. Images are stored as pixel values in the array.
There are some major differences between RGB and grayscale images. In a gray
image, there is only one parameter for color. The syntax is (height, width, 1). One can
neglect the third parameter, as it is 1. Then it could be represented as (height, or
width). Moreover, it is visualized as a 2D array. The RGB image comprises red,
green, and blue dimensions. The syntax is (height, width, 3). The third parameter in
the syntax corresponds to the color channel. The colored image is stored as a 3D
array.
4.10 Convolutional Neural Networks

A convolutional neural network is a branch of a neural network that can handle
computer vision tasks successfully. They are used in recognizing objects, faces,
scenes, and many other applications.
An artificial neural network layer consists of multiple layers in which the nodes are
interconnected. Each layer accepts the input performs some computations and
transforms it and passes the result to the next layer. This is how a standard artificial
neural network works. CNN could be considered as a particular section of the
network. This section contains filters that then contribute to pattern recognition.
The ability to recognize objects comes naturally to human beings. However, this
object identification does not come so easily to computers. This is where the
Convolutional neural network (CNN) appears. CNN is a branch of deep learning and
advances in pattern recognition. It is a feed-forward network. Deep learning
recognizes an object in an image using CNN.
4.10.1 Layers used in CNN for image recognition

 Input Layer
It takes the input in the form of pixels and stores it in arrays.
 Hidden layers
These layers perform some computations and extract features from the image in
multiple ways so that the data could be readable. The hidden layer comprises the
convolutional layer, ReLu layer, and Pooling layer.
 Convolution Layer
Convolution is a mathematical operation that combines two signals. It could learn
parameters from the data. This helps the model to find feature extractors that could
139
accomplish a task. If we imagine an 8X8 image or image of any other size then in the
traditional neural network, each one of the hidden units is connected to all pixels in
the input. If we think of an RGB image of 300x300 size then it will have 27000
weights for a single neuron. This establishes many connections.
If we continue to build such models that have fully connected units at each layer then
the model will result in a huge model with low processing speed and may come across
the problem of overfitting. To avoid this we connect each neuron to a local region of
the input value. The second factor that is considered is if one feature is useful in one
part of the input then it could also be utilized in other parts. Thus, same weights can
be shared across the inputs. Thus, convolution is performed in this unit.
A Convolution layer consists of convolution units. The parameters that are learned by
each of these units in a convolution layer could be thought of as a filter. The outputs
are the filtered versions of the inputs, which are passed to the activation function like
ReLu. Convolution layers use less number of parameters where the same parameters
are used in more than one place. This increases the statistical and computational
efficiency of the model.
This layer applies filters in the form of a matrix to the image and performs
convolution to discover patterns in the image. Convolution means coiling or twisting.
Thus, we twist and turn the data in different ways to gain several patterns from the
input image. The convolution operation separates CNN from other neural networks.
The convolution function forms the base of any CNN.
This network converts an image into pixel values and is represented in the form of an
array. The image pixel matrix is created. A kernel/filter of a specific size is applied to
the matrix for an image. The filter matrix will slide over the image and the dot
product is calculated for each element. The summation of the products results in a
single value which is then stored in the feature cell. Many filters are applied to the
image, which helps in extracting different features. Thus, we can identify patterns
from the input. The output of this layer is the feature map.
 Filter/ Kernel
The filter is applied in the form of a small matrix. The size of the filter should be less
than the input image’s size. Kernel and Feature detector are the alternate names used
for filters. The convolution layer could be made up of several filters, which are similar
in size. The size and number of filters should be pre-decided. Each filter is added to
140
the layer to satisfy a specific objective. CNN learns certain properties of the images
from the dataset during the training phase. The filter configuration is determined by
the components inside the filter.
 Stride:
The stride is a key element that decides how many pixels the filter will move over the
image. The filter moves from left to right and top to bottom. If we move the kernel by
one pixel the stride=1, if we move it by two pixels then the stride=2. The size of the
feature extracted depends on the value of the stride. The type of extracted feature also
depends on stride.
 Feature Map
Convolution operations are executed on part of the image using a particular filter. The
output of this is stored in a feature map. The number of attributes in the feature map is
equal to the number of sub-sections of the image. The feature map serves as input to
the pooling layer.
 ReLu Layer
The rectified linear activation function is a linear function. If the input is, a positive
value it reflects 1 otherwise it puts zero. For most neural networks, it is a default
function. By using this function, it is easier to train the model and achieve better
performance. This layer considers the feature maps obtained from multiple filters and
considers their weights. Several ReLu layers are taken into account for getting better
feature locations on the input. Rectified feature map is what we get as an output of
this layer.
 Pooling Layer
The Relu layer goes through the Pooling operation. The Pooling layer works on
dimensionality reduction. There are two types of pooling operations namely Max
pooling and average pooling. In Max pooling, the maximum value from the feature
map is taken. It is better than average pooling as max pooling reduces noise.
 Flattening
The flattening process converts two-dimensional arrays from pooled feature maps into
a single long continuous linear vector.
 Padding
The image reduces in size with the implication of each convolution layer on it. To
eliminate this problem padding is applied. It is a simple process of adding a layer of
141
zeros on all four borders of the input image. In general, if an (f*f) filter is applied to
an (n*n) image then changes occur in the dimensions of the image when a
convolution operation is performed on the image.
Fig 4.4: Formula of Convolution Operation on the Image

 Dense layer
These are the completely connected layers. They work on evaluating the final output
from the image. There could be multiple dense layers in the network model. These
layers are responsible to differentiate and place the data in an appropriate category.
4.11 Transfer Learning

Transfer learning helps to build deep learning models with a fast and easy approach. It
is popular in computer vision-related projects. A lot of data and computational
resources are required to develop a model. However, with transfer learning, we can
pick up any off-the-shelf module and use it to solve a problem or design a new tool.
For example, if there is a module that is used to classify objects then modifying
certain parameters could be used for identifying scenes. Studies show that features
learned from very large image sets like ImageNet could be transferred to various
image recognition models using the transfer learning methodology.
There are different ways in which we can transfer knowledge. One way is to remove
the top layer of a pre-trained model and initialize a new layer. The top layer
parameters will then be trained. The other parameters remain fixed. Thus, this
generates a system with a traditional layer at the top, which is fully connected, and the
layers having fixed parameters work like a feature extractor. This is useful if the data
and the processes in the original model are similar. If there is a lack of data to train a
model then also one can use transfer learning. It also solves the problem of
overfitting.
4.12 Model Creation
4.12.1 Developing an emotion recognition model from facial expressions:

In the research work, a neural network model is developed. It is trained on the facial
expression dataset. The trained model will then be used to predict the emotions of the
user. The aim is to understand the behavior of the students with the support of the
142
emotional analysis that we could do with the help of this model. There are many
datasets available on the internet with neurological readings and sensor-based
readings, which could be useful for emotion detection. However, here the focus of the
study is on facial expressions as they are captured through the camera.
4.12.2 Dataset used in the model

Feb-2013 facial expression dataset is used. It is available on the Kaggle website. The
data is made of grayscale facial images measuring 48x48 pixels. The faces are
registered through an automated system in the dataset. Each face occupies almost the
same space in the image. Each face is roughly centered at the same location. The
dataset has seven fields representing emotions like anger, disgust, fear, happiness,
sadness, surprise, and neutral. Two major folders in the dataset are the train and test
datasets. The training data has 28,709 examples of images. The test data consists of
3,589 images. The train folder contains seven subfolders. Each of these this subfolder
contains the images for the respective emotion. Similarly, in the test folder, there are
seven subfolders having images of respective emotions.
Kaggle:
It is a very big platform used by data science practitioners all over the globe. They
post any problem or challenge that they are facing on this platform. Experts all over
the world try to resolve it and come up with best solutions. In this way, a
brainstorming technique is used by kaggle. This is how deep learning is
revolutionizing day by day. Kaggle hosts several competitions and FER 2013 was a
part of the competition.
FER 2013 Dataset used for the facial expression analysis and classification
In 2013 Ian Good fellow along with his colleagues presented a paper, “Challenges in
Representation Learning: A Report on Three Machine-Learning Contest”, wherein he
described three different datasets. Face emotion recognition was one of this dataset. It
is well-known database for facial emotion identification. This FER-2013 dataset was
generated by fetching the images from the videos or photos posted on the internet.
There are many videos on YouTube and similar channels, which could be acquired
and used. Video is nothing but the continuous streaming of images. These set of
images from different video sources were used to detect faces. Thus, the public data
143
that was available was widely used for creating the dataset. However, there are some
shortcomings of this dataset. These are discussed below.
1. Imbalance problem
The FER 2013 data is divided into seven categories representing seven basic emotions
as shown in the table below. The problem here was, the expressions that were
obtained from the images could not be equally distributed. As we can see in FER
dataset, there are some angry and some happy samples. If we train a deep learning
model with such a dataset then there are chances that the trained model will be biased
towards the “Happy” emotion. To resolve this issue one needs to make all the
categories of data of equal length. Thus, this issue of dissimilarity in the size of data
in different categories is called as imbalance problem.
Data augmentation is the solution for this problem. Augmentation means to add.
Thus, scaling, rotating, translating are some of the techniques, which are applied to
the images to increase the number of images. The other method is to make use of
generative adversarial networks. These networks help to regenerate images with some
amount of training.
2. Intra-class variation
The images that were gathered for a particular class of emotion showed variation. For
example, the expression could be classified as angry but the image could be a photo,
sketch, a cartoon image or a painting. This may lead the deep learning architecture
into the problem of overfitting wherein the model works perfectly fine on training
data but does not give accurate results once it is deployed. Thus, the solution is to
avoid overfitting.
3. Occlusion detection
Occlusion refers to the images where the features of the face are covered with
something. For example if a person is, crying and covering his/her eyes with a
handkerchief .Another example could be during COVID everyone was using mask
due to which his or her nose and mouth were covered. It becomes difficult for the
model to recognize the emotions if the facial features are covered. Thus, it is
necessary to eliminate such type of images from the training data.
Similarly, eyes are very important for understanding the true emotions. If someone is
wearing sunglasses or any other type of eyeglasses, it again becomes difficult for the
deep learning algorithm to catch up the exact emotion of the person from the image.
To detect emotion of surprise, the eye expressions are very helpful. Also sometimes,
144
there is a thin line of difference between certain emotions or there may be images
reflecting mixed emotions. In such cases, eye expression becomes the unique factor
that helps in classifying and identifying the dominant emotion.
4. Contrast Variation
There is a range of grayscale images. There could be certain images, which are more
towards the darker side of the scale or towards the extreme brighter side of the image.
Though the emotions expressed are same, the contrast factor may hinder the emotion
recognition.
5. Outliers
There are exceptions everywhere and we come across it in the dataset. There may be
certain images, which were not belonging to a respective dataset but were included
due to automatic cropping. Such images are deliberately kept in the database. Such
images put forward a challenge to the deep learning architecture.
4.12.3 Data pre-processing

Pre-processing operation is implied on the images. Image data generators are created.
One for training and the other for testing purposes. In the trained generator, the “flow
from directory” tool was used. This is a very effective tool provided by Keras. The
tool moves through all the sub-directories present within the train directory. It will
then fetch the images from each of these sub-folders.
It will label the images with their respective emotions. The target folder and the target
size of the image are provided to the image generator. The target size of the image is
48x48 and the batch size is 64. The grayscale images will be loaded in the color mode
option and they would be classified into several categories. Thus, the class mode is
categorical. The same structure is passed to the validation generator/ test generator.
After this step, the training and testing data are transformed into processed data,
which could now be used in the model.
4.12.4 Convolution neural network developed using Keras

Keras provides two types of application interfaces that are sequential and functional.
To develop this module, the sequential model was accessed from Keras and then the
number of layers were added to it. As the layers were added in sequence, it is named a
sequential model. It is simple but a rigid model. It does not accommodate sub-layers
and can work only for a single input-output pair within the network. Developing a
145
CNN model is analogous to a sequential approach as every layer has one input and
output.
Fig 4.5: CNN Model

Classes used in the model
1. Sequential ( ) class
The sequential model is built using this class. The layers are added serially by using
the add ( ) method.
2. Conv2D () class
This class is used to create each layer of the model. It enables convolution operations
to take place in two-dimensional space. To realize the layer creation we need to pass
values to some parameters such as filters, kernel_size, strides, padding, activation, and
input_ shape. These parameters are explained below.
a. Filter
One has to decide the number of filters to be applied. It is also referenced as the depth
of the feature map. It is an integer value. In the model, the filter value eventually
increases from 32, 64, 96, 128, and so on.
b. Kernel_size
It sets the kernel window’s height and width. A pair of integers such as (3, 3) is used.
Typically, the height and width of the window are equal.
c. Strides
It is an integer value. The default value is (1, 1). This element helps to decide how
many pixels the filter will move over the input image.
d. Padding
“Valid” and “same” are the two choices from which one has to select to implement
padding. If one does not want the padding to be applied to the image then the “valid”
option is selected. Otherwise, “same” is opted. When stride=1, “same” cause padding
with zeros so that the feature map size is equal to the input image’s size.
e. Activation
146
There is again a choice on the kind of activation function to be used. In the model, the
“relu” activation function was used in each layer. No activation functions such as
linear or identity are the default option.
f. Input_shape
It is an integer number that gives the height, width, and depth of the input image. It is
only included in the first layer of the model which is immediately followed by the
input.
3. MaxPooling2D( ) class
This class is used to design the pooling layer. The convolution and pooling layers
execute as a pair in the model. The values for the parameters like pool_size, strides,
and padding are passed through this class.
a. Pool_size
It determines the size of the pooling window. The default value is (2, 2).
b. Strides
The default value is “none”. Strides define how many pixels the pooling window
should move on the feature map.
4. Dense ( ) class
In Keras, the fully connected layers are formed using the dense () class. Three
parameters used in this class are below:
a. Unit It specifies the number of nodes in the layer by passing a positive integer
value.
b. Activation
The default activation is “none”.
5. Flatten ( ) class
In between the last pooling layer and the first dense layer, there exists a flattened
layer. The input data for the MLP component of a CNN is stored in the flattened
layer.
BatchNormalization ( )
This technique helps to speed up the training process of the model. When the model is
to be trained for a large dataset with numerous features then it is very helpful. In
addition, the accuracy of the model is regulated in the validation phase. This layer
helps the model reach the optimum values quickly. In less number of epochs, the
network could get better accuracy. In the designed model, it is added before the
activation function. It brings the primary data to a standard format. For this, it is
147
important to mention the input shape argument. Similarly, to standardize the output of
the hidden layer batch normalization could be included. Normalization process turns
the input in the range of zero to one. While standardization makes the mean value
zero and the variance will be one. [169]
4.12.5 Softmax function

It is an activation function used in the output layers of the Convolutional neural
network. It is used for multiclass classification problems. The CNN layers produce
unprocessed outputs. The softmax activation function works on this output and
transforms it into a probability distribution over the n input classes. This function
generates a vector of probability values and saves the order of the scores. For
example, if the vector generated by the softmax, function has values (0.6, 0.3, 0.1)
then there is a 60 % chance that the output belongs to class A, a 30% chance of the
test input belongs to class B and 10% chance of it belonging to class C. Thus, the
class label corresponding to the highest probability value will be assigned to the test
input. The formula of Softmax function is as given below [167]:
Fig 4.6: Formula of Softmax Function [167]
148
Fig 4.7: Explanation of parameters in the Softmax Function [167]
4.12.6 Model Compilation

Conversion of Categorical data into Numeric data
The images are classified as per the emotions and are labeled accordingly. The model
however cannot put the tested input into a particular category by reading the labels.
For this, the data is to be converted into numeric values. There are two methods by
which this could be done.
1. Integer encoding/ Label Encoding
Generally, it is applied when the categories are of ranking scale like “poor”,
“average”, “good”,” better”, and “best”. The ordinality could also be considered and
weights could be assigned.
2. One hot Encoding
In this type of encoding, “1” will represent the label of the predicted output and for
the remaining class labels; it will be “0”. Thus if there are three classes and the output
belongs to the first class label then one hot encoding will be displayed as [1,0,0].
149
Moreover, the selection of the encoding method depends upon the definition of the
loss function.
4.12.7 Categorical Cross entropy

Cross entropy is an optimization function. When the model has to be trained for a
multi-class classification problem then one can make use of this loss function. As
mentioned above one-hot encoding values are represented as 0 or 1 and are allocated
to the output label. Keras.utilsis used to convert the integer output label into
categorical encoding. [168]
4.12.8 Adam optimizer

Optimizers could be considered an enlarged class. It is essential to find the right
optimizer as it increases the training and efficiency of the deep learning model. The
optimizer and the loss function work together to achieve better performance. The
model parameters are adjusted according to the output of the loss function. The
weights and learning rates are adjusted to minimize the loss. An adaptive moment
estimation (Adam) optimizer is used in the emotion detection model. It is based on the
stochastic gradient descent method. It adds some portion of the earlier gradient to the
existing one. It works on the idea of momentum and estimates the first and second-
order momentum. This is widely used while training neural networks in various
applications.
By adding portions of earlier gradients to the current one, Adam uses the idea of
momentum, which is widely used while training neural networks in various
applications. [170]
4.12.9 Metrics
The accuracy class of metric counts the number of times predicted values match the
labels.
4.12.10 Model Summary

A 2D Convolution layer, the Maxpooling layer, was added. To avoid overfitting drop
out of ( ) was added.
The convolution model was then compiled using a loss function optimizer
150
Fig 4.8: Model Summary
151
4.12.11 Model Fit
It is the measure of the level of accuracy with which the machine-learning model
generalizes that data, which is similar to the data that was used for training the model.
The outcome of well-fitted models is generally quite accurate. The one for over-fitted
model is very close to the desired outcomes, while the under-fitted model may not
produce ‘As Desired’ outcomes.
Epoch
This parameter decides the number of iterations to be performed over the data. While
training the network it is very crucial to decide the number of epochs. As number of
iterations lead to the problem of overfitting. If the number of epochs is more, then the
model gives good results and accurate predictions for the sample dataset but fails to
perform well on new data values. Hence, it is necessary to explore the optimum value
for the epoch. Generally, the data is divided into training and testing. This testing data
is supervised to get an appropriate epoch number. [172]
Shuffle
It rearranges the dataset used for training after every epoch. It is a Boolean value.
4.12.12 Saving the Model

Model checkpoint
It is merged with the model fit () to save the weights in a file called checkpoint file.
The model is saved to preserve the best weights. A Boolean value is passed. If it is
set to true, optimal results are saved. The state of the model is saved and it could be
called back later if the training is to be resumed.
Verbose
It displays the status of training. It is an integer value of either 0, 1, or 2. 0 stands for
silent mode, progress is not displayed. 1 displays the training state through a progress
bar and 2 highlights the line-by-line epoch status.
Batch_size
It is specified as an integer number. The default value is 32. For the new model, the
batch size assigned is 100.
4.12.13 Steps used in video handling

1. The dataset FER2013.csv is originally in train and test format so there is no need
for any conversions.
152
2. The data is arranged in two columns. X is the input column for pixel. Y is the
output column containing emotions.
3. Perform data pre-processing on the X_train, X_test inputs, and y_train, y_test
outputs.
4. The data is trained using the developed CNN model.
5. The model is saved in JSON format.
6. The web camera captures the image.
7. Once we get an image, the model needs to detect the face from the image and then
the emotions. Thus, Emotion detection is a two-step process.
8. Here, the haarcascade_frontalface_default.xml was used to detect the face from
the image that was captured from the webcam.
9. Resize the shape of the image so that it then becomes compatible with the
captured image from the webcam.
10. In the next step, the resized image is applied to our model and the emotion of the
image is predicted.
11. We get a sequence of images from the video. The images are captured at regular
intervals from continuous video streaming which helps in emotion prediction at
regular intervals.
The FER 2013 dataset, data pre-processing, and the CNN model have been explained
above. After the model creation, it is applied to the images captured from the live
streaming and there comes the role of the haar-cascade algorithm which is detailed
below.
4.13 Haar Cascade Algorithm

The algorithm is widely used in face and object identification within images as well as
in real-time videos. Face recognition is becoming common in many applications.
Laptops, iPhones, attendance systems, proctor-based examinations, cyber security,
policing, and monitoring are some of the areas where it is used frequently. The
“Viola-Jones Face Detection technique” is the other name for Haar Cascade.
Viola and Jones published a research paper in 2001 “Rapid Object Detection using a
Boosted Cascade of Simple Features”. In this research, the algorithm is used to edge
or line detection features. Many positive images containing faces were used to train
the model, also those images, which did not have any faces, were given for training.
These models could be accessed from the open GitHub repository. The classifier
153
models for frontal face detection, eye detection, and upper body and lower body
detection are available. [171]
4.13.1 Identification of Haar-Wavelet/ Haar Features

During face detection exercise, it is very important to detect facial features like eyes,
nose, lips, forehead, eyebrows, and cheeks. Haar wavelet, which is also called as Haar
feature, is a pattern of square shaped functions that allows us to detect these features.
It quite similar to Fourier analysis. Alfred Haar proposed it in 1909. The Haar features
are very much similar to convolution kernels. These are important features for
identifying the important features of a human face or object in an image.
It is assigned to every single feature on the human face. In a grayscale image of a
human being, we can observe the peculiarities of these facial features. We can
observe that there is a pattern of brighter and darker areas adjacent to each other. If
we consider eyebrows, then the eyebrow area is darker as compared to the upper
forehead area, which is very bright. Similarly, if we see Nose and Lips, then we see
pattern of dark-bright-dark areas. Such patterns are generalized and are called as Edge
and Line features. The Edge feature is a combination of dark and bright features
together either in horizontal or vertical directions. While, the Line feature is a
combination of dark-bright-dark or bright-dark-bright areas again in horizontal or
vertical directions.
Thus, we can see that a Haar Edge feature can represent the facial feature of eyebrow.
Similarly, the facial features Nose and Lips can be considered to be representing Haar
Line feature. These features have white and black pixels. In real images, one cannot
find black or white pixels exclusively. In real-time images, one comes across
grayscale images where each pixel is in the range of 0-255. Thus, we can allocate
Haar features to the most relevant features on the face of a human being. It is
important because the algorithm can identify humans in an image based on the
relative features. This can be well understood by the images given below:
154
Fig 4.9: Haar Features
Fig 4.10: Representation of facial features with Haar Features
4.13.2 Mathematical Computation for Haar-Features

As, we saw above Haar features help in pointing out the lines, edges and regions
where there is a sharp shift in the pixel intensities. The Haar features detect edges and
lines in different regions. A number having an intensity can represent each pixel in an
image. The lowest intensity represents white pixels and the highest ones represent
black pixels. In an ideal Haar feature, the pixels could be white or black shown with 0
or 1 respectively as shown in the image below:
155
Fig 4.11: Ideal Haar Feature’s Pixel Values
However, when we are dealing with gray-scale images in real life the pixel intensities
differ. As explained in the matrix below 0.1 is brighter, whereas 0.3 is a bit dark, 0.6
is even darker and 0.9 is almost black. The closer the value towards 0 is considered
bright pixel and the value closer to 1 is considered dark one. Thus, every image is a
combination of such dark and bright pixel value. This can be viewed with following
image:
Fig 4.10: Realistic Sample Haar Feature’s Pixel Values

Viola jones's algorithm is used to compare each image with respect to the image
having ideal Haar feature. This is done using following computational steps:
Computational steps
1. Perform the addition of the white pixel intensities and calculate the average.
2. Add the black pixel intensities and take the average.
3. Delta between these areas is then computed as subtraction of average of white
from average of dark values i.e. Average Dark Intensities – Average White
Intensities
The computation is explained below with an example.
The delta value for the ideal case is computed as “the sum and average of all black
pixels, which is one – the addition and average of all white pixels, which is zero”.
Thus, the delta value comes out as “1”. For the real image, the average of the darker
pixels is 0.8 [(0.9+0.9+0.8+0.6) / 4] and for lighter pixels, it is 0.2 [(0.1+0.2+0.2+0.3)
/ 4]. Thus, the delta value was calculated as 0.6. If the Haar value is close to one then
it is interpreted that the haar feature in the form of an edge or line is present in the
image. Likewise, horizontal and vertical edges are also found. For this, the complete
image is scanned. We will never get “0” or “1” as these are threshold values.
156
4.13.3 Integral Image in Haar Cascade
To find the presence of a haar feature one has to sum up many values. The image is
made up of many pixels and not just eight pixels. Thus, it becomes a computationally
tedious task if one has to calculate the delta value. This problem leads to the concept
of integral images. The problem is that we need to calculate the average of a given
region several times. It has the time complexity of the O (N2). The integral image
approach helps us in achieving a running time complexity of O (1). Many operations
need to be carried out because there are varied image sizes and locations. Viola
Jones's algorithm uses kernels with different sizes. The number of features could
approximately go up to 200,000. Thus, every time it is not feasible to use a quadratic
algorithm. Therefore, the integral image approach provides an optimal solution to the
problem.
4.13.4 Computing an Integral Image

The original image is transformed into an integral image. The pixel value in a given
integral image is the sum of pixel values to the left of it and above it. While
calculating the similarity to the haar feature there is a need to add the pixel intensity
values for a given region. In the integral image approach, the addition of all the pixel
values is not required. For example, if we pick the bottom right pixel of a particular
region then the value stored in the cell is the sum of the values
In the given example, the integral image concept is explained. Instead of using all the
values of the rectangle, we just use four values. Four pixels on the integral image are
used. Therefore, the sum in the integral image is: Sum = 3.7-0.5+0.2-1.7=1.7
4.14 Adaboosting
There could be more than 160,000 feature values within a detector at 24*24 base
resolution, which needs to be calculated. Adaboost helps in finding the most relevant
features from a given dataset. This helps to identify a face. The relevant features help
to represent the edge feature of the nose. It will lead to high values for this feature of
the image. Whereas, in the other image one could not get any related feature as the
upper part of the lip does not show any prominent change. Thus, it will be treated as
an irrelevant feature. The Adaboost algorithm figures out a certain number of features.
After this weight is given to these features. The linear weighted combination of all
these features helps in evaluating and deciding if the face is present or not.
157
4.15 Weak classifier
A weak classifier is always better than a random guess. For example, if we give a
hundred images to the algorithm then it will be able to detect at least fifty images
having faces. This is because AdaBoost extracts a relevant feature. The algorithm
continuously combines all these relevant features with their corresponding weight to
get a strong classifier. The output of the weak classifier is either one or zero. The
output is one when the algorithm performed well and identified the feature when
applied to the image. If the bridge of the nose feature is applied and if it is detected
then we say that the classifier has passed and will give one as output. Otherwise, if the
pattern is not present in the input image it will output the value as zero. These weak
classifiers combine to give a strong classifier. Generally, 2500 weak classifiers are
used to form a strong classifier.
4.16 Cascading
Every time we need to calculate the twenty-five hundred features that are obtained
after performing an AdaBoost. If we have an input image of 640*480 resolution then
one needs to move this 24*24 window across the image and in every single window
we need to evaluate the 2500 features and also take a linear combination and find the
outputs and see if it exceeds a specific threshold or not and then decide if the face
exists or not. Generally, a setup of the same hierarchy is used to detect if it is face or
not. The concept of the cascade was introduced where out of the 2500 features the
first ten features were added in one classifier. The next twenty-thirty features were
added in the second classifier and around hundred features in the third classifier than
the next set of features in another classifier and so on.
Thus, the complexity goes on increasing. Therefore, a cascading classifier is
composed of stages each containing a strong classifier. Therefore, all the features are
grouped into several stages and each stage has several features. The advantage of this
approach is that the hierarchy is applied to the input image and it is checked if the
input passes through the first stage of the hierarchy. Similarly, at each stage, the input
is tested. However, in case the input does not pass through the first classifier then it
gets eliminated there itself. So, no need to send it through further stages. In real-time,
this method is very useful to detect regions of the window with non-faces.
4.17 Applying the facial expression model on the live video streaming
158
The trained model is stored inmodel_from_json file and the learning weights are
stored in the model_weights_file. We load this model and check how it works on the
images captured from the live video streaming. Thus, we move toward the testing
phase of the model.
4.18 Workflow while implementing the trained model:

1. Import the required packages. Here, OpenCV was used to capture the live streams
from the video. Numpy for numerical operations and Keras model to load the
model, which was stored in JSON file and weight, file.
2. A class with the name Facial Expression Model was created and an emotion list
containing all the seven emotions” Angry”, “Disgust”, “Fear”, “Happy”, “Sad”,
“Surprise”, and “Neutral” was passed to it. The output could then be directly
mapped with the label. For example, “0” represented the angry emotion. Likewise,
the emotions were indexed from “0” to “6”.
3. The trained model structure was then loaded from a JSON file and stored in the
instance called loaded_model_json.
4. In this step, the model_from_json was used to convert whatever has been loaded
in the JSON model. Thus, here the “self.loaded_model” was created.
5. Now the weights or learning from the model_weight_file was passed to the
“self.loaded_model_weights”.
6. Therefore, to summarize until now, the convolution neural network, which was
stored in the JSON file, was created back with the help of model_from_json. Then
the self-model that was generated was given the weights.
7. Now the emotion model is loaded and ready to do the predictions.
8. The emotion model was then given data from the video. For this, define get_data.
get_data gets data from the “Video Capture” object and classified them into a
face or no face. Here the “haarcascade_frontalface_default.xml” was implied.
9. The images detected were converted to grayscale. As the model was trained on
grayscale images and was able to detect emotions in a better way on grayscale
images.
10. The face region was cropped from the image then it was given to the emotion
detection model as input for prediction.
11. Sometimes there could be more than one face identified in the video.
159
12. Each face is detected by “face.detectMultiScale”. It provides the x, y, w, and h
positions. Here “x”, and “y” is the starting position and w and h represent the
width and height respectively. This helped to draw a rectangle around each face.
13. The above step helped to gain the region of interest “roi” from the image. Thus,
all the grayscale images were stored in “fc”.
14. The emotion detection model was trained on a size of 48*48 and grayscale. Now
that the image was converted to grayscale, the next step was to resize the image.
15. The preprocessing steps were applied to the images we got from the video and
then passed to the model for prediction.
16. In the outcome, we get a list of emotions and the percentage of confidence the
new model has for the emotions was reflected. From this, we get maximum
emotional confidence.
17. The OpenCV “putText” was used to pass the emotion index to the emotion list
that was generated in the initial steps.
18. The predicted emotion was displayed in text format on the rectangle above the
face. The coordinates for positioning the emotional text were also provided.
19. Thus on the screen, we got the faces with rectangles and respective emotions
displayed on it.
20. In the last step, all the resources that were used were released.
4.19 Summary
The second module of the research project was thoroughly discussed in this chapter.
The facial expression analysis, the face recognition algorithms, were the part of the
chapter. Computer vision, image processing, and the importance of emotional analysis
in teaching-learning environment were explained. The emotions of the students were
monitored through live video streaming, which was described in the chapter. The
chapter comprises of the working of the Deep learning model CNN.
160
CHAPTER 5
THE STUDENT SURVEY APPLICATION
5.1 Introduction
In the prevailing chapters, we reviewed various difficulties faced by the teachers,
students and the entire academic fraternity in handling the COVID pandemic,
especially on the learning front. It was observed that the pandemic has brought about
many changes in the physical, behavioral, inspirational, spiritual, and various other
facets of the student’s personalities while attending the online learning mode. In
addition, these factors proved to be instrumental in managing the pandemic situation
in a varied way by each student.
The researcher tried to focus on the physical, behavioral, motivational, co-curricular
and extracurricular changes that took place among the students during the pandemic.
The research work tried to evaluate the positive, negative and neutral impacts that the
pandemic left on the students.
As this research work is deductive and quantitative in nature, after thorough analysis
and exploring multiple options, which will be discussed in this chapter subsequently,
it was decided to go ahead with the development of the Students’ Survey Application.
In this chapter, we will visit the application used to create the survey, collect the data
and perform analysis based on the collected data to come up with the result. We will
also see various frameworks that were initially explored to select the most appropriate
one for this research need. This includes study of different tools and techniques
inherently used by these frameworks, like the programming language to create a web
based application.
In this chapter, we will go through the details of this specific application and the
associated methods used to get the desired outcome. The researcher focused on
evaluating the psychological facets of the students, and thus the application is based
on the machine-learning (ML) and deep learning (DL) algorithms. For the successful
implementation of an ML algorithm, precise and comprehensive data collection was
very critical.
Thus, the researcher conducted the survey for students with ages ranging between 10
to 23 years studying across multiple educational institutions. To fetch the data from
the students whom the researcher could meet in person, the Students Survey
161
Application was used. The researcher decided to use a secured and online survey
application to get the data remotely. In both methods, a questionnaire was devised and
shared with the students. The detailed explanation about it is given in the second half
of the chapter. At the end, the results were displayed in a concise graphical format.
5.2 Basics of a Web application

A web application essentially has two classifications i.e. Front End and Back End
development. Front End represents the visual part of a web application with which the
user interacts i.e. the user interface (UI) and the command line interface (CLI)
including menus, texts, images, videos etc. While, the Back End represents the
functional and logical part of the web application with which it is built. It also
includes saving of the data in the database and the applications interaction with the
database for fetching the data as per the need. These aspects are done at the backdrop
of the application and are not exposed to the user. In short, Front End is ‘What user
sees’ while Back End is “How it works’.
Although, the application built by the researcher has both the aspects, the basis of
selection of the framework was largely driven by the convenience of Back End
development. Typically, the languages used to develop Front End are HTML, CSS
and JavaScript, while that used for Back End are Java, Python, PHP, Node.js
Let us now see the exploratory process through which the framework has been
selected for this research work:
5.3 Elements of framework selection
5.3.1 Programming languages

The programming language to be used in the development of the application based on
a framework for the research project plays an important factor for deciding it. The
learning curve, simplicity, knowledge capitalization of the language are some
important factors that plays crucial part for it’s selection. The researcher has explored
frameworks based on three programming languages that are popular for the web
application development and those are JavaScript, Ruby and Python. Following is the
brief description of them:
5.3.1.1 HTML
162
HTML means HyperText Markup Language. It is the standard for displaying the
documents in web browsers. In this, the HyperText refers to the link between the
source and destination pages while the Markup is the tagged based structural language
to define and provide meaning to the HTML document. Various browsers use HTML
to interact and work with different user interface elements like text, images, videos &
other relevant data to display in the desired format. Typically, HTML is used to
display the content in a static web page. However, it can also embed scripting
languages like JavaScript that enhances the functional behavior of a web page making
it more dynamic. HTML can also work with CSS to enhance the appearance of the
web pages.
Fig. 5.1 Structure of HTML
5.3.1.2 CSS:
CSS is an abbreviation of Cascading Style Sheets, which is a stylesheet based
language used to provide much enhanced appearance to the web pages as compared to
plain HTML. It works with documents that are created using HTML or XML based
markup languages. It provides various ways with which the elements can be rendered
in the visual display, in audio, video and other forms of content.
CSS has become the standard for all the browsers and is treated as one of the key
languages used in Free Web concept. Earlier, the development of various modules of
163
CSS used to happen in a synchronous mode. That allowed creating a single version of
CSS. It continued to create these single versions until CSS2.1. Post that there has not
been a unified version of CSS specification.
Post CSS 2.1, the scope of CSS kept on increasing and the development started with
an approach similar to micro-services. Which means, each module or component of
CSS is independently developed and maintained. Also, unlike earlier way of releasing
single version, now the most stable CSS specification is being released while the
development of various modules progress on their own pace.
The format or style of CSS used inside HTML document is as given below:
Fig. 5.2 Structure of CSS
With the example mentioned in the image above, all the “h2” type selectors in the
HTML document will be applied with the color as red and font size of 14 pixels.
5.3.1.3 JavaScript
JavaScript (JS) is a lightweight, high-level and interpreted language. It runs in single
thread and operates on all the platforms. It is one of the most popular languages and
also known as a scripting language for web development. It is primarily a weakly
typed language that can be used for both client and server side application
development. Its applications include development of websites, web applications,
server side development using Node.js, gaming, mobile applications, machine-
learning etc. Lately, it has become very popular along with HTML and CSS to create
dynamic content for website development, which provides objects controlling the
browser and Document Object Model (DOM). It also supplies the objects to run the
JS on server side and it uses Node.JS as a framework. Although, JS has spread its
usage across the globe with its power and efficiency, it has limitations like higher
security risks, weaker performance, higher-level of complexity in learning &
implementing and weaker error handling. The coding structure of JS is explained in
the image below:
164
Fig. 5.3 Coding Structure of JavaScript [185]
5.3.1.4 Ruby
Ruby is an interpreted, easy to learn, object oriented, high level and an open source
language. It can easily be embedded inside HTML and used for server side
development. It works on Linux and Windows platforms. Ruby may appear to be
difficult in debugging for some developers. The features of Ruby are explained with
the following image:
165
Fig. 5.4 Features of Ruby [186]
5.3.1.5 Python
Python is a high level, general purpose and probably the most widely used language.
It is an interpreted language used for server side web application development. It has
become very popular in Artificial Intelligence (AI) and Machine-Learning (ML) field
due to its strength with powerful mathematical libraries. It can be used on multiple
platforms including Linux, Mac, and Windows etc. It may seem to be comparatively
slower having little weaker memory management to other languages. However, it is
considered to be one of the best languages for mathematically and computationally
intensive applications. Let us see the coding structure of this language in the image
below:
166
Fig. 5.5 Coding Structure of Python [187]
5.3.2 Frameworks
Usage of frameworks makes the life of a developer easy as these provide inbuilt,
reusable, generic and commonly used tools, libraries and other components. These
components help the developers to automate certain aspects of web development and
build the websites or web applications at much faster pace reducing the coding efforts.
The types of frameworks can be Front End or Back End. In this research project, we
would be referring the term ‘framework’ mainly in the context of backend web
application development. For this research project, the researcher compared following
five such frameworks.
5.3.2.1 JavaScript based Node.js

Node.js is a free (open-source) JavaScript runtime environment and not exactly
classified as framework that is available across almost all the platforms. It is largely
used for the real time web applications like gaming, apps for internet of things, tools
for communication etc. Some of its pros include scalability, easy to learn, uniform
runtime environment across front end and back end with full stack, higher
performance. Some of the cons include higher security risks, may not support
extensive scalability, limitation on supporting synchronous & computationally
intensive tasks.
167
5.3.2.2 Ruby based Ruby on Rails
It is a server side web application framework based on Ruby. It is developed with an
intent to build the applications faster. The changes within the code for highly complex
applications can be easily managed, as it is easy to learn and implement.
Maintainability and Extensibility are its major pros along with higher level of
security. The cons include lower performance during runtime, lack of flexibility, and
comparatively lesser support for quantitative (mathematics) intensive applications.
5.3.2.3 Python based Django

It is a python based back end web application framework, which is free and open
source. It has higher level of affiliation towards computationally intensive data based
applications and machine-learning applications. It is also one of the highly popular
Rapid Application Development framework that allows fast development from
concept to implementation phase. One of the most critical advantages of working with
Django is its ability to provide higher level of security and cushioning against
vulnerabilities. It also provides desired level of scalability as per demand even during
heavy web traffic. One more differentiator compared to other frameworks is it gels
well with the social networks especially on content management systems front.
5.3.2.4 Python based Flask

It is also referred as micro framework as it does not require any specific library or tool
for web development. The most common functionalities are provided as third party
libraries, which are made available along with the framework. The intent of coming
up with this framework is to develop simple applications with less features and
developed in easy and faster mode. Some of the pros with this framework are the
applications developed with Flask are easily scalable, easy to build and implement,
higher level of efficiency & performance and simple in nature. The cons include its
nonstandard way of implementation across companies, which makes it difficult for
knowledge capitalization and sharing; higher time required for deployment of the
solution and comparatively slower performance than other python based frameworks.
5.3.2.5 Python based FastAPI

This is one of the most recent and new age Python based framework, which supports
developing APIs. The intent with which it was built was quick development, lesser
defects during the software development lifecycle and higher level of runtime
168
performance. It completely supports the asynchronous mode of coding. Some of its
pros are it has an inbuilt autocomplete function that helps quick development that also
helps in efficient debugging. It helps in datatype validation coming from JSON. On
the cons front, FastAPI needs everything to be tied together with the application, thus
the main file in the application becomes very bulky. In addition, this is relatively very
new framework, thus the community of using it is very less. Thus, getting technical
support on this framework is not so encouraging including the knowledge
capitalization.
5.4 Framework Selection

As seen above, the researcher did thorough comparative analysis across multiple
technologies based on JavaScript, Ruby and Python. The parameters used to decide
the technology to implement for this research work were:
 Easy user interaction and transition between pages / tabs
 Easy to learn and implement
 Quick development
 Affiliation towards Machine-Learning
 Scalability as per the need of the research project
 Data security
 Maintainability
 Easy integration of video application to capture facial expressions
 Easy connection and extraction of data from database
Based on the above parameters, Node.js, Ruby on Rails and three python based
frameworks were studied. Some points were slightly better in Node.js and Ruby on
Rails, however the most prominent requirements of this research project were getting
better satisfied with the Python based framework. Later, within Python based
frameworks, the researcher further did the comparative study between Django, Flask
and FastAPI. As we see from the points mentioned above, though Flask is better on
flexibility and FastAPI is slightly faster than the other two, Django having higher
level of packages, higher level of security and robustness and strong connection with
database SQLite, the clear selection for the research project was Django.
Let us now take a deep dive into the Django framework for back end web
development application as mentioned below:
169
5.5 Django Framework
The python framework called “Django” makes it simpler to develop websites using
python. Django helps in managing the challenging tasks of the project so that one can
focus on the creation of web applications. Don’t Repeat Yourself or DRY, is a
philosophy that promotes the reuse of components. Django includes ready-to-use
capabilities like a login system, database connection, and CRUD activities (Create
Read Update Delete). To work with Django it is necessary to have python installed in
the system. Python 3.7.4 version was used in the application designed for the research
work.
This framework enables the user to easily build effective online applications. Django
has built-in features for Admin Interface and the default database. A certain set of
elements are usually required while establishing a website. For instance a method for
handling user authentication like login, logout, user validation, and verification.
Ready-to-use components are made available by Django, which helps in faster
development.
It has a sizeable library and functions like web scraping, machine-learning, image
processing, and scientific computing. All these could be integrated well with a web
application to perform numerous advance tasks.
5.5.1 Virtual Environment

Each Django project should have its virtual environment and in the python module,
“venv” command was used to manage virtual environments.
5.5.2 Django app

An application in Django helps to create a home page, a form for collecting contacts,
or a form for registering members.
Advantages of using Django application
1. The Django apps could be reutilized across different projects.
2. Independent components exist.
3. Numerous developers can work on different components.
4. Organizing code and debugging are simple.
5. It has built-in functionality, such as admin pages, that makes it easier to create
things from scratch.
5.5.3 Django Structure
170
Model-View-Template (MVT) architecture is the foundation of Django. It is a
software design pattern for creating web applications.
The MVT structure is composed of three basic elements as mentioned below:
1. Model
The model serves as the data’s interface. It maintains data. It is represented by a
database and serves as the logical data structure for the entire application. The Django
model is analogous to the table in the database. In Django, data is generated in
objects, which are considered models.
To do a variety of tasks, such as creating, updating, deleting, or retrieving the model
fields, we can use Django’s admin interface. The Django model offers consistency,
version control, simplicity, and sophisticated information handling.
2. View
The user interface appears when a webpage is accessed through the browser. HTML,
CSS, and JavaScript files are used to represent it. A view function is a python
function that accepts a web request and delivers a web response, according to the
Django documentation.
3. Template
In Django, a template is explicitly written in an .html file using JavaScript, CSS,
HTML. A template is considered the most crucial component of the MVT structure.
The Django framework processes and produces dynamic HTML web pages that are
visible to end users efficiently.
Templates are generally used to create a front-end layout for our website because
Django mostly functions at the back-end.
5.5.4 URLs
In Django, views are used as python methods that take a URL request as a parameter
and either return an HTTP response or throw an error. Each view needs to be
associated with a certain URL pattern. This is achieved by using the URLConfPython
package (URL Configuration).
5.6 The Design of the Student Survey application

The student survey application opens with a registration form. The student login is
done with a valid username and password.
171
Fig. 5.6 Flow Chart of Student Survey and Results
The working of the student survey application is described in the flowchart above.
The student has to register by filling in preliminary information. After Registration,
the student can log in to the system. To solve the questionnaire, an application was
designed using the Django framework. The questionnaire collected the age of the
students with which we can categorize the students into “Elementary” (below 12
years), “Junior High School” (13-15 years), “Senior High School” (16-18 years), and
“Undergraduate” and “Postgraduate” (above 18 years). The questionnaire was split to
collect responses from five major categories that are Physical, Motivational,
172
Behavioral, Co-curricular and Extracurricular facets of the student. This
questionnaire has 30 queries.
5.7 Five major parameters of student’s personality considered for study
5.7.1 Physical Changes

The pandemic disturbed the daily routine of the children. School/College going
children have small and long breaks for breakfast and lunch. In addition, they were
habitual with having meals in the stipulated time and with their friends. Children were
not able to play outdoor games. All this affected the physical growth of the children.
The major factors that were noticed were lack of exercise, changes in eating and
sleeping habits. Either they started overeating or there was lack of appetite. Physical
lethargy indirectly affected the mind, emotion and behavior of the children. The tool
that was developed by the researcher tried to capture the changes on various physical
parameters. In addition, it was interesting to find these factors were correlated and
affected the overall personality.
5.7.2 Motivational Changes

Learning was shifted to online mode. We should be thankful to the technology as it at
least generated some kind of student engagement. Again, there were issues like
student monitoring. Students were exposed to whole lot of stuff on the digital media.
Children started using lot of social media platforms. On the other hand, there were
students who became techno-savvy and explored the positive-side of technology.
They started learning new programming languages and enhanced their skills. A study
was conducted to find if the children were motivated enough to attend the online
classes or they were negatively affected or were neutral to the situation.
5.7.3 Behavioral changes

The continuous exposure to online learning tools, online submissions, lack of
knowledge of handling online technology tools were certain factors that led to poor
performance in students. This further led to lack of interest amongst some students to
study online. In some students, it also led to nervousness, anxiety and depression. The
study covers teenage students. During this period, they undergo hormonal changes
and find themselves comfortable with their friends. However, lockdown and social
distancing restricted the offline gathering. This led to mood swings and aggression in
173
some children. Thus, the study tried to capture the temperament of the student,
thereby analyzing the effect on the behavior.
5.7.4 Effect on Co-curricular activities

The co-curricular activities are necessary to flourish the academic skills of the
students. Seminars, group discussion, debates, essay writing, programming and
project competitions all are the part of these activities. These were important to
develop the team building spirit, leadership skills, public speaking skills among
students. They help to inculcate reading, writing skills and upgrade the soft skills of
the students. All these activities were also shifted online. The students participated in
these activities. However, again the question was about their wholehearted
participation. In this survey, the student responses were fetched to understand if they
enjoyed participation in online activities, or they were disinterested.
5.7.5 Effect on extracurricular activities

The students are always enthusiastic in participating in these activities. Almost every
student is interested in some hobby. It is important on the part of schools and colleges
to support and motivate these students. The students play various indoor and outdoor
sports, participate in culture activities. When they represent their school or college at
University, state or district level, they get encouragement. All these activities got a
setback due to pandemic. Still it was interesting to know if students continued with
their passion or not. The questions in the survey were designed to get answers to these
questions.
The techniques used to display the results in graphical representations in the form of
charts, graphs, tables. This makes the data interpretation easy.
5.8 Python Language

Python is a programming language widely used in different domains. It is useful in
developing both standalone and dynamic applications. Python has a large collection of
standard libraries. Due to this, it is used in building Machine-learning applications,
GUI applications, and designing Web-based frameworks like Django. It is also used
in Image processing like OpenCV, text processing, and similar applications. Python
programs are generally small and easily readable. Some popular companies like
Facebook, Amazon, Google, Instagram, and Uber use python language extensively.
174
Python provides optimized solutions that eventually boost developer productivity.
Some major attributes of Python are listed below:
1. Python is a very high-level language
2. It is a general purpose
3. Supports object-oriented programming
4. Highly Interpretable
5. Widely Portable
6. Freely available
7. It is highly coherent
5.8.1 Python’s Development Cycle

Python’s development cycle does not have compilation or linking steps. The programs
written in python language use objects from modules that are imported at runtime.
This leads to the quick execution of programs even after changes are made to them.
Python supports the reloading of modules. It is even possible to modify the parts of an
executing program without stopping it. For example, the code written for a button
placed on a GUI could be easily updated and the changes could be immediately
observed on an active GUI itself. Thus, the need to stop and replay the code is
eliminated. This leads to a shorter development cycle. The figure below shows the
difference between the traditional cycle and Python’s development cycle.
Fig 5.7: Difference between Traditional Cycle and Python Development Cycle
[190]
175
Python supports incremental development. Modules are tested independently and then
integrated to form a system. The complete system is again tested to ensure proper
execution of the entire code.
Python programming provides a development cycle that combines the benefits of
prototyping, rapid application development, and hybrid application. Python language
helps in realizing these concepts into modular development without compromising on
the programming capabilities.
Fig 5.8: App Delivery Process using Python [190]
5.9 Python Libraries
5.9.1 TensorFlow
TensorFlow provides an end-to-end platform for Machine-learning and deep learning
tools. It provides a comprehensive set of tools for entrepreneurs, researchers, and
developers. It helps them to build machine-learning applications. TensorFlow
includes a good set of APIs, which helps to develop the models easily.
Google developed TensorFlow and maintains it. It was made available under the
Apache 2.0 open-source license. [176]
Features of TensorFlow
1. Features like data preprocessing, data injection, model evaluation, visualization,
and deployment facilitate better model development.
2. TensorFlow is highly portable; hence, one can train and deploy the model
anywhere and on a variety of devices.
176
3. It can be scaled from a single CPU (Central Processing Unit) to a GPU (Graphics
Processing Unit) to a cluster of GPUs and a multimode TPU (Tensor Processing
Unit) infrastructure.
4. For the quick calculation of numerous mathematical equations, it combines
computational algebra and optimization techniques.
5. It consists of many machine-learning libraries.
6. It has structured documentation.
Despite having access to the underlying C++ API, the API is very native for the
Python programming language. [177]
5.9.2 Tensor
A tensor is a vector or matrix that has several dimensions. Tensors are represented as
n-dimensional arrays of fundamental datatypes.
Shape
The Shape in the TensorFlow Python library is equivalent to the tensor’s
dimensionality. TensorFlow infers the shape automatically as it builds the graph.
Data flow and operations in the form of a directed graph are used to explain
computation. The data flow graph is comprised of nodes and edges.
Nodes
Nodes are computational units. They have zero or many inputs and outputs. Data
moves between nodes are known as tensors. These are the multi-dimensional arrays of
real numbers.
Edges
The flow of data within a graph is represented by an edge. It also defines branching,
looping, and updates to a state. Special edges are employed to synchronize tasks such
as waiting for computation on some inputs to finish.
Operation
The ability to accept input attributes and produce output attributes is referred to as an
“operation”. The operation such as add or multiply needs to be specified. [178] [179]
5.9.3 Keras
177
The basic structure of Keras helps to build TensorFlow or Theano-based deep
learning models. Deep learning models are quickly defined with Keras. Keras is a
python-based open-source neural network library on top of TensorFlow. A Google
engineer named Francois Chollet created it. Thus, Keras is a high-level API that
facilitates the development of neural networks.
5.9.4 Backend
Keras uses a low-level library called “Backend”. This library handles all the lower-
level computations like creating tensors, variables, or graphs that are not supported by
Keras. It also develops convolution with the help of TensorFlow. “TensorFlow “is by
default the backend engine used in the development of models but one can change the
configurations.
Keras supports multiple backbends. The performance of the neural network depends
on the Keras backend. TensorFlow is an open-source platform developed by Google
and released in November 2015. Another popular backend platform is Theano. It was
developed at the University of Montreal in 2007.
5.9.5 Pickle
The Pickle library in Python allows the object structure in the Python language to
convert into a byte stream and in similar way, converting back to object structure. The
process of converting object structure to byte stream is called as ‘Pickling’, while the
process with which the byte stream is converted back into Python’s “Object
Structure” is called as “Unpickling”. This process is well depicted with the following
image:
Fig 5.9: Pickle Library Overview

178
The process of pickling and unpickling is also referred as serializing and de-
serializing respectively.
Significance of Pickle:
While working on real world challenges or problems, one need to keep on transferring
the data from one system to another and in the process store this data in a file. Thus,
the ‘Pickle’ feature provided by Python is quite handy to achieve it.
Pros of Pickling:
 Can work with complex object structure
 Easy for usage, learning and implementing requiring lesser lines of code
 The output file coming after pickling is not easily readable / interpretable, thus
inherently it provides some level of data security
Cons of Pickling:
There is a security risk while unpickling in case the pickled file is coming from some
malicious sources
5.9.6 Numpy
Numpy (Numerical Python) is a free source python library. It is used in all branches
of research and engineering. It is a standard for manipulating numerical data in
python. Numpy API is substantially used in sci-kit learn, Pandas, matplotlib, and
SciPy. It is also used in many data science and scientific packages.
Data structures for multidimensional arrays and matrices are available in the NumPy
library. These data structures ensure efficient calculations with arrays and matrices.
Mathematical calculations are performed on arrays using NumPy. It provides a vast
library of high-level mathematical functions that work on arrays and matrices.
Python has lists that function like an array. However, the lists require more processing
time. A Numpy array object is 50 times faster than the conventional python lists.
Numpy arrays are stored in a single continuous location in memory. This helps
programmers to access and manipulate them. [173][180]
179
Fig 5.10: NumPy Library Overview [191]
5.9.7 Sci-Kit learn

SKlearn or Scikit-learn is a machine-learning python library. It is open-source and
freely available. It is BSD licensed hence it has a minimum complexity. It is built
above Numpy, Scipy, and matplotlib. It provides tools that could be used for data
mining and analyzing data. It consists of scientific and statistical modeling tools like
Scipy and NumPy. These tools could be efficiently utilized in classification,
clustering, regression, and dimensionality reduction. They are used for personal as
well as commercial purposes.
Sci-Kit learn supports various machine-learning algorithms like support vector
machine, random forests, k-means, and gradient boosting, etc. The Scipy toolkit
supports these algorithms. It also implements various data pre-processing, data
visualization, and cross-validation algorithms. [181]
180
Fig 5.11: Sci-Kit Learn Library Overview [192]
5.9.8 Pandas
Wes McKinney came up with the name “Pandas” in the year 2008. It refers to both
“Panel data” and “Python Data Analysis”. Pandas is the widely used open-source
library in python. It is used to manipulate data sets. Pandas have efficient data
analysis and data visualization tools. It also offers tools for data cleaning, data
exploration, and data manipulation.
Fig 5.12: Pandas Library Overview [193]

Numpy and Matplotlib are the two sub-libraries of pandas. Numpy is used for
mathematical calculations whereas matplotlib is used for graph plotting and data
visualization. Pandas came up with strong Data analysis tools as described below.
1. Series
181
Series was introduced in pandas, which is similar to a list structure. Pandas series
stores data in column format. The data is passed to the series and in the output; the
series is printed along with the label, which is termed an index. Series supports
different types of data like integers, float, strings, etc.
2. Dataframe
Dataframe was another structure that helped to store data in table format. It is a two-
dimensional data structure. Data is arranged in rows and columns, which are
considered the main parts of the pandas' data frame.
3. Matplotlib
It is said, “A picture is worth a thousand words”. Some things could be learned in a
better manner by looking at them rather than describing them. Data visualization is a
very important aspect to gain a proper understanding of data.
Python’s Matplotlib toolkit facilitates a complete tool for building static, animated,
and interactive visualizations. It is a graph plotting library used in python. It is used to
plot two-dimensional and three-dimensional graphs. Data could be very well
represented by plotting histograms, Bar charts, Pie charts, Line charts, and scatter
plots through this library. [182]
4. Pyplot
The Pyplot package consists of a MATLAB-like user interface. The pyplot function
helps to modify a figure in some way. For example, drawing a figure, plotting a
specific region within the plotting area, and labeling the plot. Image, 3D plot,
Contour, and polar plot are some of the plots that can be drawn using Pyplot. [183]
5.10 Keras Application

Pre-trained models for deep neural networks are offered by the Keras applications
module. Prediction, and feature extraction, are performed using Keras models. Model
architecture and model weights are the two major components of the pre-trained
model. There is a variety of pre-trained models available out of which VGG16 was
used in the pre-trained model for the research work.
Advantages of Keras
1. Provides Quick and easy development of network models.
2. Keras could be trained on single or multiple GPUs.
3. One can choose different back-ends for various projects.
4. Keras helps in easy Model deployment as it supports some devices or platforms.
182
5.10.1 VGG 16
VGG 16 is one of the popular computer vision models. It is a variant of the
convolution neural network (CNN) model. It is an algorithm, which can detect and
classify objects. A type of artificial neural network, ConvNet is an abbreviated name
for convolutional neural network. A convolutional neural network is made up of
multiple layers like input, output and large number of hidden layers. Thus far, VGG16
is considered one of the highest rated computer vision models, which is also a
derivation of CNN i.e. Convolutional Neural Network. The programming community
working on this model did in depth study of the networks to augment the deepness of
this model using amazingly concise convolutional filters (3 3). These filters displayed
significant progress compared to the highly progressive setups. This depth of layers
was later enhanced to produce as many as 138 parameters used for training. [184]
5.11 The working of the Student Survey application
5.11.1 Registration and Login Process

The Django based application tool was designed such that the students’ could take the
multi-page survey and at the end could get an analysis about the COVID effect on
different attributes of their personality. For that, the students had to first register and
take the survey. As we start the application, it opens following landing login page:
Fig 5.13: Landing Login Page
It can be seen that on this simple landing page, there are barely four user interface
elements and those are ‘Username’ and ‘Password’ text boxes and ‘Login’ and ‘Sign
Up’ buttons. Let us see the functions of these elements in following section:
183
Username: This has to be an unused and unique identifier for the student, which will
be used by him / her subsequently in all interactions with the application. Most
commonly used usernames are their own names. This field had been designed to
accept only alphanumeric values and some special characters like ‘@’, “_” etc.
Password: To keep the data secure and confidential, the password field is designed.
This field accepts any character that can be entered through a keyboard, the only
criteria is that it cannot be just numbers. After accepting the password from the user,
it is encrypted by the application and like any other application it is masked so that it
is not visible to any onlooker. Each username has corresponding password and this
list is internally maintained by the application, which is referred each time a student
connects with the application.
Login: The Login button simply initiates the survey and takes the student to next
page. It is available to the students who have already registered themselves on this
application before.
Sign Up: The Sign Up button would be used by the students who would be taking the
survey for the first time. Clicking on this button would take them to a form that would
require additional information to create their Login accounts inside the application as
given in the image below:
Fig 5.14: Sign Up Page
In the ‘Sign Up’ form, Username and Password fields are similar to as mentioned in
the ‘Landing Login Page’. The only additional field added to this form is:
Password Confirmation: The text entered in this field is compared with the text
entered in ‘Password’ field and if they match then it will allow taking the process of
registration ahead.
184
With the above process, the registration process would get over for the students and
now they are ready to take the survey. Let us see the generic flow of the application to
take the survey and thereafter, we will go through the detailed use case to see the
responses and their meanings in details in subsequent pages:
5.11.2 The Student Survey (Generic Flow)

The survey has been divided into several sections to gain insight about the various
parameters of the study undertaken.
The first section of the survey has tried to capture the general view of the students
about the pandemic. The student fills his / her name, school / college name and then
provides their feedback to the questions. The questions were framed to analyse the
general view of the students about their online learning experience. In addition, how
they would react if they come across similar situation in future. In this part, the
students responded to five questions.
On the top right corner of every page a “Start” button is provided to capture the video
streaming, while the students were filling the survey. This was done to get a better
analysis of the live emotions of students while they were responding to the questions.
The next page of the survey, contain questions related to physical changes that the
students’ experienced. Some students were keen on exercise and fitness while some
became lazy and lethargic. Appetite of the children was affected and there were
changes observed in the sleep patterns and sleeping hours. These and a few other
parameters were captured through the set of physical questionnaires. Whether the
child was infected by COVID was also the major factor that was checked in this
section. The students’ answered six questions regarding the physical changes that they
had gone through.
The third page gathers responses related to the motivation of the students. Factors like
the time they spent on social media, television, was considered. In addition, if they
were comfortable with the online learning option and they were asked to rate online
learning. The time they spent on online classes and self-study were taken into
consideration. All these answers were used to find out the level of motivation they
experienced. Total six questions were framed for this part of the study.
The fourth page captured the behavioural patterns of the students. It was important to
know what they feel about continuous exposure to online tools. In addition, students
had to bear long screen time. Especially this was a matter of concern for school
185
children. The loss of physical connect with the teacher also left an impact on the
students. It was crucial to understand the students’ view about online assignment
submission. The huge volume of assignments was also a point of concern for some
students.
Attempt was made to understand their views about offline education. Questions were
posed to find out if the students feel that face-to-face education helps in developing
the personality of the students and how they feel about their academic performance.
The behaviour/temperament is very difficult to understand. The students considered in
the study, mostly belonged to an age group where they come across physiological
changes. The teenage students experience some physical changes in them. On the top
of that, they had to face the lock-down and had to follow the measures laid down by
the pandemic. This all affected their temperament. The children were not able to see
and meet their friends or close relatives for long. While, some of them were connected
with their friends and relatives through online platforms. The question was also asked
in this regard to the students and the responses were collected to understand the level
of satisfaction the children were getting through such type of online social gatherings.
In all, eleven questions were asked to find the changes in behaviour of the children.
The fifth page of the application gathers the student responses regarding their co-
curricular and extracurricular activities. It was observed in the pandemic, that there
were some students who developed new hobbies, or improvised their skill and talent
by exploring the online platform related to their domain of interest. In addition, some
students find it interesting to participate in quizzes, online seminars, or group
discussions.
On the other hand, there were students who were very much active in co-curricular or
extracurricular activities in the offline mode. However, they lost interest when these
activities were conducted online. There was again a third category of students who
remained neutral in this aspect. Two questions were posed in this section to know the
impact on co-curricular and extracurricular activities.
On all the pages of the survey application, the camera was switched ON to capture the
emotions of the students while filling out the survey through their facial expressions.
After completing the survey, the student clicks on the “I’m done” button and the
application moves to the next page and answers to the questions are saved.
On this page, the user could see the message “Survey submitted”. On this page, one
has an option either to go back and change responses or can click on “Next’ and go to
186
another page. Then, on this page, the “Result” button is displayed. On the click of the
Result button, again the camera is activated for a moment to capture the emotions.
This is done because due to some reason if the student/user forgot to click on the
“Start” button for activating the camera option on every page then at least on the
result display page, the facial expression could be captured. After clicking on the
“Result”, the next page displays the physical, motivational, and behavioural impact on
the individual student. The analysis of the facial expressions is also displayed. The
impact factors are evaluated as “Positive”, “Negative” and “Neutral”.
On this page, there is a “Display Result” button. On clicking this the bar chart for the
results of survey application as well as jotform results are displayed.
5.11.3 Specific Use Case of Student Survey
Any registered student can login to the application using his / her credentials and start
taking the survey. To do that the student has to follow the steps mentioned below:
 Click the “Login” button to hook onto following page:
5.15 Welcome Page of Survey
 This would open the “Welcome to the Survey App” page having “Take the
Student Survey” button. Click on this button to open following page:
187
5.16 Survey: Personal Details
As can be seen in the above page, there are several User Interface elements. The
functionality of each of these elements is explained below:
 As mentioned in the header “Student Survey”, this survey is intended for the
students who were operating in an Online of Distance learning mode due to the
lockdown and schools operating in remote mode.
 On the top right side of the page, we can see the “Start” button that accesses the
video camera of the laptop and displays the live feed of the student’s video in the
frame just below it. In the specific case above, you can see the frame captured of
the student from her video stream. Note that before starting this feature, it pops a
message to allow the access to the laptop’s video; the user has to say “Yes” to
capture the video.
 The name of the form is “COVID Lockdown” with a foot note of “About COVID
Lockdown”
 There is a main frame on the form “Answer this part” embedding five more user
interface elements in it.
 The first element is “What is your Name?” expecting the name of the student.
 Next field named “School Name / College Name” requires the student to provide
his / her school name.
188
 Next field took the input from the students based on their experience with the
online learning approach. The response was to be provided out of the following
five options
o Very Good
Leading to a positive response
o Good
o Neutral Leading to neither positive nor negative response
o Bad
Leading to a negative response
o Very Bad
 The field after that took the feedback about their health while the lockdown was in
force. Here too there are five responses to the question as mentioned in the above
point with same meaning for each set of response(s).
 The next question is very important to check the experience of students about their
attendance during the online classes having same set of responses and meaning.
The response to this question is considered vital because it gives a trend of how
many students have taken the online experience as positive and how many of them
have taken it in the negative way. In addition, we also would get to see the facial
expressions through the frame captured in the video section. We can quickly
verify the response and check the correlation with the facial expression.
 Lastly, this form has “Next” button, clicking on which takes the survey ahead on
to the page with next set of questions.
189
5.17 Survey: Physical Change Analysis
The students define the above page to check the “Physical” Changes that might have
observed while learning remotely. It has following set of questions:
 First question on the form takes age in years as an input from the student
 Further, it asks for the time in hours that is spent by the student on taking care of
physical health through some fitness means. No or minimal time spent would
mean “Negative” while optimal time spent means “Positive”.
 Third question takes the response about the “Number of meals taken per day” by
the student during the pandemic times.
 Fourth question checks for the “Change in weight” as an important factor to see
the change in the physical health. There are 3 choices given to the students to
choose from:
o Decreased Leading to a positive response
o No Change Leading to neither positive nor negative response
o Increased Leading to a negative response
 Subsequent question determines the number of hours of sleep that the student had
per day. Optimal number like 7 or 8 hours leads to “Positive” feedback while very
less like < 6 hours or higher like > 9 hours is considered as “Negative” feedback.
190
 The ultimate question explicitly asks if the student had faced any challenges in
terms of health having options to choose as “Yes” or “No”. While “Yes” would
lead to “Negative” response and “No” leading to “Positive” response.
 Answering the questions for above fields would end the questions on this page
and clicking on the “Next” button would take us to following page:
5.18 Survey: Motivational Change Analysis
This page takes the input from the student about the “Motivational” changes observed
in them during the Online Learning phase. This form had following specific questions
to analyse the inputs coming from students:
 The response to the first question seeks to know about the time in hours spent by
students connecting to social media. The higher the number of hours spent, it
would categorize the response as negative and lower number means positive.
 Similar to the above question, the next question about spending time (in hours),
and watching television would carry the same meaning and categorization i.e.
higher the time spent, its negative and lesser is positive.
 The next question tries to get the response about student’s time spent in hours for
the self-study. As the question is posed in positive way, the higher the number, it
would be treated as positive and lower the number, it would be treated negative.
191
 Like the earlier question, the next question seeks the time spent by students during
the online classes. It was observed that some students being sincere could attend
for entire time, while some students were not inclined to attend all the classes.
More attendance in the class indicates positive response and less attendance
indicates negative response.
 The question after this checks with the students about how was their time spent
during online classes. It has two responses – Yes and No. The answer “Yes”
would be considered to be positive while the answer “No” would be considered as
negative.
 The last question in this form asks the student to rate their overall experience with
the online learning mode as they attend the classes.
o Excellent
o Good
o Neutral Leading to neither positive nor negative response
o Poor
o Very Poor
 The ‘Next’ button would take the student to the next page with another set of
questions.
192
5.19 Survey: Behavioral Change Analysis
This page having highest number of questions tries to deal with the “Behavioural”
part of the students’ experience. All the questions had same five set of responses as
given below, to determine the behavioural level of students.
 Strongly Agree
 Agree
 Neutral Leading to neither positive nor negative response
 Disagree
 Strongly Disagree
Let us see the questions and the feedbacks in details in the following section:
 Initial question checks with the students on their agreement or disagreement about
their consistent exposure to the screen, which might have led to the tiredness and
exhaustion in them.
 Next it determines their view about in person learning through schools, colleges or
universities giving rise to their enhanced social profile and confidence.
 Third question tries to identify students’ feedback if they were at home for longer
time, does it cause boredom and lethargy in them.
 Subsequent query checks if prolonged use of eLearning tools generate anxiety and
stress within students.
 Next question talks about the economic disparity within society with a question
whether some students are deprived of digital tools due to lack of money for them
and has it led frustration for them.
 Further, it tries to understand the opinion of the students about their
recommendation of discouraging the usage of digital platform for learning from
longer term perspective as it could not be healthy socially as well as
psychologically.
 Another question in the list tries to see the impact of lockdown, closures and
quarantine phases posed by COVID-19 pandemic on the stress levels of the
students and to check if it had led to depression for some students.
 The second last question asks whether the huge number of assignments had
caused confusion and lowered performance.
193
 The last question tries to seek the inputs about the importance of face-to-face
interactions of students with each other and their other friends, which helps them
to improve their academic achievements.
 Answering above questions and then clicking on ‘Next’ button would take us to
the following form:
5.20 Survey: Co-Curricular and Extracurricular Change Analysis
This form checks the interest or passion of the students towards co-curricular or
extracurricular activities those were conducted in an online mode. It has following
couple of questions to determine it having two responses – ‘Yes’ and ‘No’. The
earlier response is categorized as positive while the later as negative:
 The first one in the list tries to seek the interest level in participating in the co-
curricular activities like online quizzes, seminars and group discussions.
 The other question checks the passion level of the students towards extracurricular
activities like hobbies.
Answering the queries for all the above-mentioned forms truly helped the researcher
to get the insights about various parameters (Physical, Behavioural, Motivational, Co-
194
Curricular, Extracurricular) pertaining to the personality of the students. With this last
form, the survey gets over and let us see how the results are displayed using this
application. On clicking ‘I’m done!’, it would take us to following page:
5.21 Survey: Submission Page
Upon receiving this message “Survey submitted”, the students can get confirmation
that their inputs have been processed. The “Back” button will then take back the
students to the first Login page. The “Next” button will take them to following page
to see the results.
5.22 Survey: Result Link Page
Navigating the “Result” button will display the results in textual format as given
below:
195
5.23 Survey: Results in Text Form
Exploring the results through “Display Results” button, will show overall effect on
students in graphical format as given below:
196
5.24 Survey: Results in Graphical Form
197
5.11.4 The Admin view
To perform certain administrator level tasks like adding user from backend, deleting a
specific user or multiple or all the users, editing responses (e.g. typo correction),
deleting a survey or many or entire set of surveys, a Superuser is created with higher
level permissions. This can be accessed using following form:
5.25 Survey: Admin Login Page

A superuser was created with username as “Admin” with a relevant password to
access the Admin related tasks. Using these credentials, one can get into following
pages:
198
5.26 Survey: Site Administration Home Page
As can been seen on the page above, a site administrator has an access to
Authentication and Authorization view of all Groups, Users and also to Survey and
Report view of all sets of answers to surveys and all surveys. It also provides a view
of all recent actions performed using the application from anywhere.
For example, clicking on the ‘Surveys’ link, we get following view with a survey
filled by a student. We can also observe that this survey can be selected and there are
some associated actions that can be performed like “Deleting the survey” through the
options provided in the drop down field “Action”.
On the right side, we can also note that a new survey can be added through backend
process using “Add Survey +” option. In addition, we can also filter the normal users
and authenticated users based on their access criteria to view the results like “Yes”,
“No” etc. The image below shows it:
5.27 Survey: Site Administration for Surveys
199
Similarly, navigating the link “Users”, it lists all the users with their details and access
status i.e. normal user or superuser. Here too, we can select one, multiple, or all the
users together to take certain actions provided in the ‘Action’ drop down box. For
example, deleting some set of users. We also have filters to select users and
superusers based on certain criteria. Following is the image depicting it:
5.28 Survey: Site Administration for Users
5.11.5 Jotform
 Jotform Basics:
Aytekin Tank in the year 2006 established “Jotform”. The company is situated in San
Fancisco. Jotform is an online, powerful form building tool. It is a completely
featured and freely available form creator. Millions of users all over the world have
developed faith in Jotform and are using it for collecting data. It is a portal to gather
data from the end users in a better way, which can then be processed and used by the
organizations. It is useful to non-profit organizations, educational institutions, small
business or enterprises. It also helps in developing a dedicated and flawless payment
collection mechanism. There are certain characteristics of jotform, which were
explored to some extent and were used while creating the Jotform for this research
work.
 Creation and working of Jotform
Jotforms are very simple and easy to generate. The form builder helps to create a form
without coding. To develop a form one needs to create an account.
Jotform account
200
One can create an account by following a simple procedure. On the home page one
gets a pop-up where one can toggle to decide if one has to sign through a Google,
Facebook or any other email-id.
Pricing options in Jotform

Starter, Bronze, Silver, and Gold are the four pricing plans offered by Jotform. Apart
from starter all other are paid plans and provide an annual billing system. The starter
is a free plan wherein the users can create five forms, could take one hundred
submissions per month, and could utilize 100 MB of space. The bronze, silver and
gold plans charges were $24, $29 and $79 per month. The form limit is 25 and 100
for bronze and silver and storage space is 10 GB and 100 GB respectively. While the
gold plan offered unlimited forms and 1 TB storage space.
The most interesting part is Jotforms offer around 50% discounts to non-profit and
educational institutes.
Jotform Templates
It provides an extensive collection of around 10,000 templates in different categories.
People from any industry could easily find template suitable for their category. These
templates are very well organized so that one can find the required template or at least
end-up finding something close to what he / she is looking for and then customize it
according to specifications. [188]
Following steps were followed to create the survey form.
STEPS:
1. Create an account and login.
2. On the dashboard, there is an option of “CREATE FORM” click on it.
3. One has to select from the four options those are “Start from scratch”, “Use
Template”, “Import Form” and “Create signable document”.
4. For the survey form, the researcher used the “Start from scratch” option.
5. On clicking the “Start from scratch“, choice is to be made to select the layout i.e.
classic form or card form.
6. The classic form layout was selected.
7. On this layout, the online survey questions were dragged and dropped.
8. On the left hand, side top corner there is an option of “Add Form Element”.
201
9. At the click of “Add Form Element” there are three categories i.e. “Basic”,
“Payments” and “Widgets”. Out of this, the “Basic” element was selected.
10. “Basic” listed out various elements like heading, full name, long, short, multiple-
choice, number, single choice, image, time, file upload, captcha, scale rating etc.
One can pick up the type of questions to be included in the Form.
11. On the top, we see “Form”. When we tap this section, we get an option to add a
header to the form. Likewise, the header and sub-header were added to the survey
form.
12. One can also select the “Heading” option from the basic element.
13. The ‘Full name” element was also selected to get the full name of the student.
14. For the research form the elements such as Short text (College / School name)
multiple-choice (experience good, average, poor), number (for age, meals, sleep
time etc.), were used while adding the questions.
15. Whenever we add any of the above elements onto the form a settings icon appears
on the right side. On clicking it, a window appears on the right side where we can
fix and assign the properties for the element that has been added to the form.
Integrating and setting the Take Photo “Widgets”

At the end of the questionnaire, the students were supposed to take photo and upload
in the form. This was done through “Widgets” in the “Add Form Element” panel.
These Steps were followed:
1. In the “Add Form element” panel click on the widget tab. Search the Take photo
widget, then drag and drop it onto the form.
2. After it is added, automatically the widget settings window will appear on the
right side of the screen.
3. If the device has more than one camera, it needs to be decided which one to be
used. For this set the “Camera Facing” Front/ Rear option. In the survey, the Front
option was set.
4. The other text properties like take photo, retake photo, notification were also set.
The notification was set with this message “Your photo has been successfully
uploaded”.
5. The survey form does not give students a choice to click photo. Hence, the
properties were set as “prevent submission if this field is empty”. Thus, the form
202
was not submitted unless the student’s photo was uploaded. As it was required for
the facial analysis of the student.
The Take photo widget eliminates the need to exclusively open the camera app on
your device. Once the user was done with the survey questions, he / she had to click
on the take photo button. Tap the camera icon to take a photo and you are done with
it. The picture immediately is uploaded on the form and was prepared for submission.
This prevents the time and effort to take the photo outside the jotform application.
Thank you Page and Publish

Once the Form building task is completed, the “Thank you” page was selected by
moving on to the settings tab. On this, there are other options like integration, email
alert configuration and conditions.
The survey form was then published by clicking on the “Publish” tab. This Tab gives
us various ways to share the form with users. The other method is to click on the
email tab so that a link is created. For the survey, the email tab was used to create
link, which was then shared.
Jotform submissions:
The student survey responses were supposed to be viewed and sorted. For this jotform
provided the submission tab. To view the submissions “My Forms” was opened.
Then, the form whose entries are to be checked was selected. Next, the submission
button in the top tool bar was clicked. After this, the user was redirected to jotform
tables. From here, the student entries were collected.
Jotform Tables:
Once the students started filling the form through link, there was a need to collect the
entries. On the top right corner of the login page there were options to download the
responses in either excel, CSV or PDF file. The excel option was used to gather and
store data. On the “My Form” tab, one can see the total number of submissions and
the latest date when the form was updated. In addition, the recent date when the
student submitted the response could be viewed from this page.
5.11.6 Using Jotform for data collection

To collect the data from the students operating in remote mode, instead of hosting the
application on cloud due to security threats and potential leakage of sensitive data, the
203
researcher created a Jotform and shared with them through a link in an email.
Following images show a specific use-case on how the data is collected through a
mobile friendly form created using Jotform:
1. To take personal information details:
5.29 Jotform: Personal Details
2. To take the inputs for Physical parameters:
204
5.30 Jotform: Physical Change Analysis
3. To take the inputs for Motivational parameters:
5.31 Jotform: Motivational Change Analysis
4. To take the inputs for Behavioral parameters:
205
206
5.32 Jotform: Behavioral Change Analysis
5. To take the overall feedback about online classes and also to capture student’s
photograph:
207
5.33 Jotform: Survey Submission
After collecting the data from Jotform, it is fed inside the Django application to
enhance the training data for the ML models.
5.12 Summary
As per the requirement of the research work, various technologies and applications
were explored and based on the suitability for this research work, a survey application
using Django framework was shortlisted for implementation and subsequently
developed. This application has been thoroughly described in the chapter. The
framework required for designing the survey was studied. Various python
208
frameworks, libraries, the lifecycle of python programming were described. The
working of the application was shown through the live example. Using this example
the complete survey could be understood. The second part of the chapter showed the
survey method. This was developed using the jotform. The structure of the forms was
detailed using a real life use case. The chapter contains the features, advantages,
applications, and limitations of the survey.
209
CHAPTER 6
RESULTS AND DISCUSSIONS
6.1 Introduction
To meet the objectives of the Research the Machine-learning and Deep learning
algorithms were applied and Models were developed. In the third chapter, four
machine-learning algorithms and their workflow were discussed. These algorithms
were used for the model creation. The fourth chapter details the CNN model creation
for the emotion analysis through facial expression recognition. In the fifth chapter,
the survey application tool was explained. The evaluation of all these models and the
survey tool are discussed in this chapter.
Four machine-learning models were implemented for three parameters of students’
personality. These models deal with the analysis of the physical, motivational and
behavioral patterns of the students. Thus, each model is used for evaluating the effect
in the above three categories. The analysis was done to find out as to how the COVID
pandemic affected the students. The study reveals the positive, negative or neutral
effect on the children.
Effect on the co-curricular and extracurricular activities were also the parameters for
the study but the models were not created for these categories. The analysis was done
through framing the questions, which were the part of the survey tool.
The second module of the research work is a deep learning model that was developed
using convolution neural network. The facial emotion analysis was done through it.
This module was designed to analyze the facial expressions of students while they
were responding to the questions of the survey application. In all twelve machine-
learning models and one deep learning model was implemented for conducting the
study. All these models were developed using Python.
The survey application was used to collect the responses of the students belonging to
several schools and colleges. The application contains self-designed questionnaire.
The results and effects of the student responses under each category are depicted
through the graphical representations.
210
6.1 Evaluation of the ML Models for Physical, Motivational and Behavioral
analysis
Once the model was created, it was necessary to train the model using available
datasets. For the entire machine-learning models, “Kaggle dataset” was used. The
physical and motivational models were trained using a single dataset. While for
exploring the behavior/temperament of the students two datasets were merged into a
single dataset. After this, the model was trained and tested. To find the appropriate
model for identifying the physical changes, motivational and behavioral changes in
the students it was important to check the accuracy of the model predictions. It was
found that all the four algorithms showed different accuracies for each of the
categories. In addition, it was interesting to observe how each of these algorithms
reflected the positive, neutral or negative effects. It was observed that some model did
not predict a particular effect in a specific category. This was crucial in determining
which model to choose for each of the specified categories.
6.1.1 Assessment of the Models

The performance of the models was key factor in deciding which models should be
included in the final assessment of the students through the survey application. The
classification reports of each model in all the three categories as well as the confusion
matrix for the models is briefed below. Precision, Recall, and F1-score were the
prominent factors for evaluating the models.
6.1.2 Confusion Matrix

The Confusion Matrix is a metric that helps to find out to what extent a classification
algorithm was successful in producing right result. It is extensively used as it provides
a more accurate picture of a model’s accuracy rather than the classification accuracy.
It is a technique to summarize the performance of the classification algorithm. The
matrix helps to determine the correct classification made by the model as well as the
errors made by it. Confusion matrix compares the predicted data with the actual data.
This helps to know if the model is a Good fit to be used. Good fit model is one, which
has high number of true positive, and true negative cases whereas less number of false
positive and false negative cases. When one has to work with unbalanced dataset, it is
preferable to use confusion matrix as an assessment tool. The performance of the
classification model can be examined by calculating Precision, Recall and F1-score.
211
There is no information on the amount of misclassified occurrences in model
accuracy.
 Precision
It is a statistical term that shows how many of the correctly anticipated instances were
truly positive. Precision is one of the most acceptable metrics to identify, when would
be the costs associated with the False Positive would be higher. For example, when
we receive an email, some of them are marked as spam. However, the categorization
of an email in spam or not spam largely depends on the precision of the algorithm. If
the precision is higher, then important mails should not be categorized into spam
(false positive case) and it is a desired case.
 Recall
It shows how many of the actual positive cases one was able to predict correctly with
the help of the model. It would be an ideal model when the cost related to false
negative is considerably higher. For example, if the truly suspicious transactions done
through online payment mode were predicted to be seamless transactions, then the
resulting scenario would be very painful for the consumers.
 F1-Score
It captures both the trends in a single value. In cases, where the balance is to be
sought between precision and recall, then F1-Score is the ideal metric.
 Accuracy
It is a metric with which we come to know how many times a model is predicting
actual or desired result. It is the ratio of number of accurate predictions to the total
number of predictions.
With the following image, the difference between Precision and Accuracy is well
explained. Precision is to predict the same result consistently, while accuracy is each
time getting closer to the actual result.
212
Fig 6.1: Relation between Precision and Accuracy [189]
6.2 Performance Interpretation of the KNN model
Analyzing effect on student Behavior
213
Fig 6.2: Confusion Matrix for Behavioral Analysis using KNN
In the above figure, the output given by the KNN model is shown in the form of the
confusion matrix in the category of behavioral analysis. It was observed from the
matrix that the KNN model for behavior, predicted the negative and neutral effect
very well. However, the model was unable to analyze the effect in the positive
category.
The model predictions were significantly high for the neutral class. The correct
predicted percentage for neutral category was 86.96%. The matrix reflected 83.33%
correct results for negative class. This was the second highest score. There were no
predictions made for the positive class.
Understanding the effect on student Motivation
214
Fig 6.3: Confusion Matrix for Motivational Analysis using KNN
The figure above, shows that the KNN model worked well in finding the level of
student’s motivation. Out of the three classes, the model could predict the neutral
effect quite well. The accuracy for the neutral class was 73.61%. In the negative
category 70.79% results were indicated correctly. Again, the KNN model was not
capable enough to estimate the student’s motivational effect in the positive category.
Estimating the effect on students’ physical changes
215
Fig 6.4: Confusion Matrix for Physical Analysis using KNN
The model depicted better results in the negative category for physical changes. It
showed 65.62% efficiency in predicting results in the negative class. The outcome in
the neutral class was 37.50%. The positive effect could be measured in physical
domain by the KNN model. Unlike the behavior or motivation class, at least the
model could make some predictions for the positive class. The model depicted very
low result that is 33.33 % in the positive effect for physical category. These effects
are presented by the Confusion matrix in the figure above.
6.3 Interpretation of the SVM model

Behavior Analysis
216
Fig 6.5: Confusion Matrix for Behavioral Analysis using SVM
The Support vector machine model displayed output for all the three positive, neutral
and negative classes. The percentage outcomes were 100%, 87.80 % and 94.12%
respectively. The highest accurate predictions were made for the positive class while
examining the behavior changes among students. The second highest predictions were
in showing the negative effect. The lowest predictions were made for the neutral
effect identification. The results are summarized in the figure above, in the form of a
confusion matrix for the behavioral analysis.
Motivational Analysis
217
Fig 6.6: Confusion Matrix for Motivational Analysis using SVM
The SVM model worked fine in measuring the motivational effect on students. It
could find only the negative and neutral effect. It could not predict the results for the
positive class. The result for the negative effect was 81.48%. It showed 81.25 %
output for estimation of neutral parameter. In the above figure, the confusion matrix
details the results gained in all the three categories.
Physical Analysis
218
Fig 6.7: Confusion Matrix for Physical Analysis using SVM
The model displayed varied results in the three classes. It got 100% correct results for
the positive class. Thus, it achieved the highest score for the positive effect. In the
neutral category, the correct predictions achieved were 50%. These were the lowest
interpretations by the model among the three categories. The negative category
showed 69.23% results. The above figure of confusion matrix for physical effect
analysis displays all these results.
6.4 Interpreting the results achieved by the NN model

Effect on student Behavior
219
Fig 6.8: Confusion Matrix for Behavioral Analysis using NN
The Neural Network Model gave the best results in the behavior category. For the
positive class it showed 75% correct analysis. In the neutral effect recognition, it
resulted in 94.74% efficiency. For the negative part, it gave quite high result of
94.96%. Thus, it showed comparatively low result analysis for the positive class, but
for the remaining two sections, it performed remarkably well. The Confusion matrix
shown in the above figure explains the results of all the three categories.
Effect on student motivation
220
Fig 6.9: Confusion Matrix for Motivational Analysis using NN
In the student, motivation category the NN model showed good results for the
negative and neutral class. For the positive label, it was able to gain 50% correct
predictions. This was the lowest among the three categories. The outcome of the
neutral class was 84.42%. This was the second highest in the motivation section. The
model for the negative attribute could make 86.84% results analysis. This was the
highest output achieved. The outcomes and effect on student motivation with respect
to the three attributes in shone through the confusion matrix in the figure above.
Effect on students’ physical Attribute
221
Fig 6.10: Confusion Matrix for Physical Analysis using NN
While studying the physical effect on student through the NN model it was observed
that in the neutral parameter analysis the NN model achieved very low results. In the
positive category, moderate outcomes were received. In the negative section, highest
output was gained. The percentage results for the neutral class were 51.92%. For the
positive effect 66.67%, results were depicted. For the negative parameter, it was
82.95%.
6.5 Interpretation and Results of the RF model

Examining the effect on student behavior
222
Fig 6.11: Confusion Matrix for Behavioral Analysis using RF
The Random forest model was able to show better results in finding the effect on
student behavior. It reflected highest score in the positive class. Negative attribute
prediction score was the second highest. The lowest outcome was for the neutral
class. The percentage predictions for the positive class were 100%. For the negative
class it was 94.12%. The output for the neutral parameter was 87.80%. The confusion
matrix in the figure above shows the results for all the classes.
Analyzing the effect on student motivation
223
Fig 6.12: Confusion Matrix for Motivational Analysis using RF
The model displayed best results in the category of student motivation analysis. It was
observed that the model could accurate predict the positive effect on student
motivation. It could very well examine the neutral attribute. For the negative
parameter, it showed bit lower accuracy as compared to the other two. The model
percentage efficiency for positive effect was 100%. It had the accuracy of 93.15% in
the neutral class. Whereas, for the negative effect evaluation it showed an accuracy of
87.21%. The figure above presents a confusion matrix, which represents the effect on
students’ motivation.
Evaluating the physical effect on students’
224
Fig 6.13: Confusion Matrix for Physical Analysis using RF
The RF model exhibited good results while evaluating all the three attributes that
were considered under the students’ physical effect. It analyzed the positive effect
with moderate efficiency. The lowest prediction rate was for the neutral category.
While the highest predictions were generated in the evaluation of the negative,
attribute.
The percentage results were 84.62% for the positive class, 82.05% for the neutral
feature analysis, and 86.46% for evaluating the results of the negative attribute. The
above figure compiles the effect on students’ physical changes in the form of a
confusion matrix.
Classification Report and Comparative Analysis of the four models
KNN Models
We have seen in detail the results of the four machine-learning algorithms through the
confusion matrix that were represented for each model in the above section. In
addition, classification report discussed above clarifies the difference between the
225
efficiency of the models. In this section, a comparative analysis is done to get a crux
bout the capabilities of each of the four models.
Category Precision Recall F1-Score
-1 0.86 1 0.93
0 0.89 0.58 0.7
1 0 0 0
Accuracy 0.87
Macro Average 0.59 0.53 0.54
Weighted Average 0.86 0.87 0.85
Table 6.1: Classification Report for Behavioral Analysis using KNN

To start with, the first KNN model for behavior is discussed here. It gave an overall
accuracy of 87%. The classification report above also reflects this. The major
drawback of this model was that it was not able to identify the positive effect at all.
This is understood from the confusion matrix displayed above.
-1 0.71 0.82 0.76
0 0.74 0.67 0.70
1 0 0 0
Accuracy 0.72

Table 6.2: Classification Report for Motivational Analysis using KNN
Then, the second model using KNN was developed for finding the effect on student
motivation. In this category, again the KNN algorithm could classify the negative and
neutral effect very well but could not give the outcome for the positive label. The
model indicated an efficiency of 72%.
226
-1 0.63 0.85 0.72
0 0.38 0.31 0.34
1 0.67 0.08 0.15
Accuracy 0.57
Table 6.3: Classification Report for Physical Analysis using KNN

The third model for physical analysis was designed using the KNN algorithm. It
expressed very low efficacy of 57%. However, it recognized the effect in all the three
labels. It worked average for the positive, negative, and neutral attributes.
If we compare the three models of KNN. The behavior and motivation models scored
high but they both missed on analyzing the positive factor. The third model for
physical changes could classify effect in all the three labels but overall efficiency was
low.
SVM Models
The support vector machine algorithm was used to design models in the behavioral,
motivation, and physical category. The confusion matrix and the classification report
for these models elaborated the applicability of these models.
-1 0.94 0.97 0.96
0 0.88 0.84 0.86
1 1 0.33 0.5
Accuracy 0.93
227
Table 6.4: Classification Report for Behavioral Analysis using SVM

The behavior model developed using SVM gave 93% accurate classifications. It
depicted very good metric scores in positive, negative as well as neutral labels.
-1 0.81 0.86 0.84
0 0.81 0.82 0.82
1 0 0 0
Accuracy 0.81
Table 6.5: Classification Report for Motivational using SVM

The second model was created for finding the effect on motivation of the students.
The model depicted 81% efficiency in classifying the neutral and the negative labels.
This model could not find the results for the positive label. Hence, it was not
appropriate for application.
-1 0.69 0.92 0.79
0 0.50 0.39 0.44
1 1.00 0.25v 0.40
Accuracy 0.66
Table 6.6: Classification Report for Physical using SVM
228
The physical model was designed using support vector machine algorithm. Its overall
efficiency was 66%. It was able to find the effect in all the three classes. It performed
very efficiently for analyzing the positive label.
Neural Network models
-1 0.95 0.98 0.97
0 0.95 0.84 0.89
1 0.75 1 0.86
Accuracy 0.94
Table 6.7: Classification Report for Behavioral using NN

All the three models that were developed using Neural Network algorithm worked
remarkably well. The behavioral model showed maximum accuracy of 94%. This was
the highest accuracy that could be achieved in comparison to the behavioral models,
which were developed using the other three algorithms, that is KNN, SVM, and RF.
-1 0.87 0.86 0.86
0 0.84 0.82 0.83
1 0.5 0.8 0.62
Accuracy 0.84
Table 6.8: Classification Report for Motivational using NN
229
The motivational model developed by using SVM reflected an accuracy of 84%. This
was the second highest score compared to the other models developed using other
algorithms for this category. However, the positive effect predictions were not
satisfactory.
-1 0.83 0.83 0.83
0 0.52 0.55 0.53
1 0.67 0.58 0.62
Accuracy 0.71
Table 6.9: Classification Report for Physical using NN

The Neural Network physical model reflected efficiency of 71%. This was also good
compared to the physical models developed using KNN and SVM. As KNN physical
model had the lowest accuracy of 57% and SVM, physical model depicted 66%
result.
Random Forest Models
-1 0.94 0.97 0.96
0 0.88 0.84 0.86
1 1 0.33 0.50
Accuracy 0.93
Table 6.10: Classification Report for Behavioral Analysis using RF
230
The random forest algorithm that was used to develop the behavior, motivation, and
physical models depicted excellent results for all the three models. The behavior
model of random forest was 93% accurate in finding the effect on the students’
temperament. The accuracy of the model was close to the Neural Network behavior
model, which was 94%.
-1 0.87 0.97 0.92
0 0.93 0.86 0.89
1 1 0.40 0.57
Accuracy 0.90
Table 6.11: Classification Report for Motivational Analysis using RF

The Random forest model for motivation could analyze the effect on motivation of
students up to an efficiency of 90%. This was the maximum output that was achieved
by the RF motivation model as compared to the other algorithms used in motivational
model development.
-1 0.86 0.94 0.9
0 0.82 0.65 0.73
1 0.85 0.92 0.88
Accuracy 0.85
Table 6.12: Classification Report for Physical Analysis using RF
231
The physical Random Forest model proved to be the best in comparison to the three
other models that were implemented. The model score was 85%. This was the highest
score that could be accomplished in this category considering the NN, SVM, and
KNN motivational models.
Best model
In all twelve models were developed using four machine-learning algorithms for three
categories. Thus, there were four models in each section that is behavior, motivation,
and physical. The four models KNN, SVM, NN, and RF for behavior analysis
exhibited the accuracy of 87%, 93%, 94%, and 93% respectively. Out of these, the
Neural Network model was the best as it gave an accuracy of 94%. Hence, this model
was considered for finding the effect on the behavior of the students.
In the analysis of the effect on motivation of the students, it was observed that the
Random forest model resulted into an efficient model. The Random forest model gave
90% accuracy. It was the highest percentage in comparison to the K-nearest neighbor,
Support vector machine, and Neural Network that represented an outcome of 72%,
81%, and 84% respectively.
The third category that was for finding the changes in physical habits of students.
Here, again the Random forest algorithm performed well with an output of 85%. The
KNN model had very low score of 57%. The support vector machine was 66%
accurate. The neural network model displayed the second highest score. It had an
efficiency of 71%.
Thus for the behavior effect analysis Neural Network model was selected. The
motivation and physical effect was analyzed using the Random Forest models for
respective categories.
Comparative Analysis
232
Fig 6.14: Comparative Analysis of ML Models
CNN model results (Camera)
Rigorous effort and analysis is required to predict the changes in the behavior of the
children. Behavior is dependent on different aspects of one’s personality. The tool that
was developed included a Convolutional neural network (CNN) model, which could
capture the emotions of the students while they were filling the survey. The emotions
were analyzed and were bifurcated into three groups positive, negative and neutral.
The FER 2013 data set was used. The model was trained using the images in the
trained dataset. The classification report shows the result of the model on the trained
dataset.
The Facial emotion recognition dataset has seven categories of emotions as shown in
the result. These are angry, disgust, fear, and happy, sad, surprise, neutral. These
emotions are further grouped and labeled. The emotions of anger, disgust, fear, sad,
and surprise are clustered and labeled as “Negative”. The emotion of Happiness is
mapped to “Positive” label. The neutral emotion was interpreted as “Neutral”.
Out of the 28,709 images used for training, the model could classify 3,995 into
“Anger” class, 436 into “Disgust”, 4,097 for “Fear”, 7,215 for “Happy”, 4,830 for
“Sad”, 3171 for “Surprise”, and 4,965 for “Neutral”.
The accuracy of prediction for “Happy” and “Neutral” class was highest. The model
accuracy rate was 99% for these two classes. Thus, the model could predict the
“Positive” and “Neutral” emotions very well.
233
The model accuracy rate for the emotions in the negative category were 98% for
anger, disgust, sadness, and surprise. The accuracy of the model was 95% for the
emotion of fear; which was somewhat low compared to other emotions of this class.
The overall accuracy of the Convolutional Neural Network model for classifying the
emotions from the facial expressions was 98%.
Category Precision Recall F1-Score Support
Angry 0.98 0.98 0.98 3995
Disgust 0.98 0.98 0.98 436
Fear 0.95 0.98 0.97 4097
Happy 0.99 0.99 0.99 7215
Sad 0.98 0.97 0.98 4830
Surprise 0.98 0.99 0.99 3171
Neutral 0.99 0.98 0.98 4965
Accuracy 0.98 28709
Macro Average 0.98 0.98 0.98 28709
Weighted Average 0.98 0.98 0.98 28709
Table 6.13: Classification Report for CNN Model
234
Fig 6.15: Confusion Matrix for CNN Model
The confusion matrix shown in figure 6.13 elaborates the percentage accuracy of the
CNN model for each class. 98.26% for Anger, 98.17% disgust, 95.22% fear, 99.44%
Happy, 98.24% Sad, 98.43% Surprise, and 98.98% Neutral.
Fig 6.16: Model accuracy and Model Loss plots

Interpretations from Student Survey application
Physical Effect
Two hundred students used the survey tool to fill the data. The impact on their
physical parameter was studied. It was found that out of 200 around 75 students
235
(43.35%) were not affected by the change imposed by the pandemic. They were
classified into the “Neutral” group. 58 students (33.53%) students reflected
disturbances in their routine activities. This contributed to “negative” impact on them.
Around 40 students (23.12%) were following physical discipline. They maintained
physical fitness in the pandemic, or there were some who even took up with good
physical activities and exercises. They maintained healthy diet and took care of
themselves. This led them in the “positive” category.
Fig 6.17: App Result for Physical Effect

Effect on Motivation
The students were not found to be highly motivated in studying in an online
environment. There were various reasons for this and there say regarding this was
recorded in the form of answers to the questionnaire. The questions for checking the
students answered the effect on student motivation and the results observed are
mentioned here. 153 (88.44%) students were neither positively or negatively
motivated hence, the “Neutral” category. 19 students (10.98%) students represented
them were badly affected and were grouped into the “Negative” category. Out of the
students who filled the survey, only one student was found to be highly motivated and
in percentage, it is 0.58%.
236
Fig 6.18: App Result for Motivational Effect
Impact on the behavior
In this category, it was observed that the students were mostly negatively affected.
There were many adjustments that the students had to make at the time of pandemic
and even after that. School students also filled the questionnaire. Mostly they are in
the adolescence phase, where they are comfortable with the peers. In their growing
years they had to face the lockdown, social distancing and some might have
undergone the isolation or other mental or physical alignments. All these factors made
them irritable, as they could not maintain mental stability because of the uncertain
time. Also their approach towards the online learning environment, tools, technology
usage. Their level of interest in using and handling all this was noted through the
survey questions. The outcomes show that 112 (64.74%) students were found to have
a bad impact on their behavior pattern. 16 students (9.25%) had no influence and 45
students (26.01%) had a good impact on them in terms of behavior.
237
Fig 6.19: App Result for Behavioral Effect
Effect on co-curricular activities
The bar chart displayed that the students were not interested in taking part in the co-
curricular activities. It was evident from the student responses that the offline
environment is suitable for such kind of activities. The students who are in the first
year of a particular course/ College are a bit apprehensive. They are not familiar with
the college, teachers. In lockdown situation, the students were facing more anxiety
and stress. They did not get a chance to visit the college physically where they had
taken admission. They were not able to make new friends. All such factors made them
somewhat reluctant to do anything extra.
The important constraint for participation was again the online mode, which kept
most of them away from such activities. Some students were not having the
knowledge or skills to participate in the activity. In physical mode, they could gain
understanding about it from their friends or seniors in the college or even from the
teachers. All these factors led to low online participation. The results show maximum
98 students (56%) were not in favor on the online conduction of the co-curricular
activities. 25 students (14.29%) students continued with the same level of
participation and had no major changes. 52 students (29.71%) students were found to
be inspired and were actively involved in such activities. Rather they enjoyed the
platform for these activities.
238
Fig 6.20: App Result for Effect on Co-Curricular Activities
Impact on Extracurricular activities
Children always feel good when they inculcate some hobbies in them. With this
objective, the extracurricular activities are promoted in the educational institutes.
These are also the source of recreation for them. This releases their stress to some
extent and helps them focus on studies in a better way. According to some students
the charm of the activities is lost if they are taken online. This was also depicted
through the survey results. It showed that 107 students (61.14%) students were
disinterested if the mode of conduction for these activities was online. It was revealed
that 36 students (20.57%) were happy and utilized the time that they got in developing
these activity skills. 32 students (18.29%) showed no change in their level of
participation. Thus, the effect was categorized as “neutral”.
239
Fig 6.21: App Result for Effect on Extracurricular Activities
The average emotion of students:
To increase the reliability of the results the emotions of the students were examined.
This was done while the students were answering the questions of the survey. It was
observed that 162 (93.10%) students were most of the time reflecting “Neutral”
emotion. This was the average emotion. Ten students (5.75%) were sad, or unhappy
hence it was represented by “Negative” emotion. Only 2 students that is 1.15% were
happy and hence were grouped in the “positive” class.
Fig 6.22: App Result for Effect on Average Emotions

Result Analysis for the student survey using Jotform
The survey was conducted for the school, Junior College, and Senior College students
residing in urban and rural areas. The student responses were captured through the
survey application as well through jotform. To fetch the data from remote areas the
jotform was framed which comprised of the same questionnaire included in the survey
application.
The data from 1,503 students was collected using the jotform link and then it was fed
to the survey application. The application then predicted the effect on the physical,
motivational, behavioral, co-curricular and extracurricular aspects.
Behavioral Analysis
The behavior of the children got affected due to the pandemic. They had to face three
to four phases of lockdown and had to handle the uncertainties prevailing around. In
the initial phases of the lockdown, the students were enjoying the stay with their
240
family members and were able to get access to education sitting at home. Later on as
the time passed, they realized the negative impact of the online methodologies. It was
stressful for students to understand the practical concepts. They were lacking hands-
on practice sessions. Also in case of any doubts related to subject, they could not
reach to the teachers easily as they could do in case of offline classes. Post that phase,
they faced the hybrid mode which was again disturbing phase for them.
In spite of all these difficulties, post COVID the students were not comfortable with
the complete offline classes as they were before. The students were habitual by now
to the use of gadgets. They just could not resume to the gadget free education as it
was before the pandemic. Definitely, COVID pandemic brought about a change in the
approach of students towards learning. Before pandemic, they were not aware of the
choice of online learning. However, pandemic introduced this option to the students.
Some students were now comfortable with online and objective type of examinations.
It was convenient for them as they lost the practice of writing subjective answer
papers.
Thus, through the survey it was found that the students wanted the best of both the
methods to be including in teaching methods. They felt that online model of education
should be made more robust and some digital methods, which could enhance
classroom teaching and experience, should be integrated with the model.
The opinion of students about the exposure to electronic devices, prolonged stay at
home, prolonged use of e-learning tools, volume of assignment submissions were
analyzed to gain insight about the impact on temperament. It was understood from the
result that most of the students had mixed responses. They agreed to 50% of the
question and disagreed or were uncertain for rest of the questions. This lead to overall
analysis as “Neutral” for behavior of students.
The graph in the figure below shows the effect on the behavior of the students. It was
found that out of 1,503 students 877 (58.35%) students reflected neutral attitude , 415
(27.61%) were reluctant to accept the change and showed negative behavior, whereas
very less number that is 211 students(14.04 %) faced the change with a positive
approach.
241
Fig 6.23: Jotform Result for Effect on Behavioral Analysis
Effect on Student Motivation
It was understood from the research and motivational theories that the college
students are mature enough and by the age of 18-19 years, they have developed a
deep-routed belief system, which is influenced, by their upbringing, socio-economic
background and their individual accomplishments and failures. Students remain
motivated when they get a chance to apply these skills in their college life. This helps
in their confidence boosting. In addition to this, they think about their learning in
terms of achieving their future objectives. This all could happen in physical model of
learning. Because of pandemic, the students were deprived of this experience and
were not able to keep themselves motivated. They lost the sense of self – efficacy.
The biggest challenge in front of the students was to remain motivated despite of the
unpredictable situations. The students suddenly got an access to the digital world.
They lost the disciplined approach that they used to follow during their school/college
hours before the pandemic.
In the traditional classes, students could remain motivated by the peer- pressure or
competition. Sometimes they could acquire a skill through team bonding or team
spirit and effort. They had lost their usual ways in which they could keep themselves
motivated.
242
There were students who started exploring the digital world and there was no check
on the amount or content of access. While some were struggling to accept this mode
of education as they were not interested in the technical assets. While some children
felt like they were on a vacation and used the social media and other facilities to
entertain themselves. There was so much uncertainty around so through the
entertainment channels children kept themselves busy.
The survey posed the questions related to the time they spent on social media,
entertainment, Television, self-study, online classes. In addition, whether the students
were happy with the online class experience and if they felt that their time was
properly utilized or not.
Students were spending long hours in watching television, or accessing social media
were obviously mapped to the negative effect category. In addition, those who spent
less time in online classes or self-study belonged to this class.
It was found from the survey conducted that most of them were not motivated to use
the online model. The bar chart in the following figure, details the effect on
motivational aspect of the students. Out of 1503 students 794 (52.83%) were affected
negatively in this parameter. Closer to this 653 (43.45%) expressed neutral emotion
and, hardly 56 (3.73%) students were positively motivated.
Fig 6.24: Jotform Result for Effect on Motivational Analysis
243
Analyzing the Physical aspect
Physical growth of the children was again an important dimension that needed
attention in the pandemic and otherwise also. Physical activities of the children were
hampered. One could observe a shift in this attribute. There were students who were
aware about the importance of the physical exercise and continued their physical
routine. While, there were some who became aware of the importance of physical
fitness and started practicing it. Therefore, pandemic turned out to be a “Boon in
Disguise “ for them. Some students rather were reluctant to follow any kind of
physical routine this could be relative to the unsureness that surrounded them.
Schoolchildren are in the growing age and require good amount of physical exercise
and proper nutrition, which would suffice their physical needs and help in proper
growth. Exercise is also a factor that influences the appetite of the children. Some
children started overeating, while others faced the problem of loss of appetite.
Before COVID, the students had a habit to eat in confined time and amount. The
students in schools or colleges would get short break for having breakfast and they
would have lunch in long breaks. However, this limitation was lost somewhere when
they were at home. Again, there were children who were infected and suffered in the
pandemic. This could have brought changes in their physic.
Research have shown that the mental health of the children also depends upon the
physical wellbeing of the students. Engagement in physical activities, healthy
behavior, and limiting screen time turned out to be the ways, to mitigate the stress.
Exercise leads to recreation of mind and body. It helps them to face and be resilient to
the stressful situations that they come across. However, the children could not play
outdoor games and had to find ways to engage themselves in physical and mental
activities.
The survey questions were projected in such a manner to gain inputs about the weight
of the students, their eating and sleeping habits. The questions were also posed to
understand their health parameter.
From the bar chart given in the figure below, one could know the results of the
student survey. 446 (29.67 %), students were positively affected in the pandemic
related to physical parameter. It was the lowest score. Whereas, 585 (38.92%)
students showed negative influence and that was the highest among the 3 categories.
472 (31.40%) students showed neutral effect.
244
Fig 6.25: Jotform Result for Effect on Physical Analysis
Effect on co-curricular activities
Every school and college puts efforts to build the soft skills, leadership qualities and
team spirit among the students through various co-curricular activities. These
activities are also important for time management and for enhancing the
organizational skills. Programing, project, essay writing, debate, group discussion,
quizzes, seminars, power point presentations are some of the examples of such
activities. They are part of the curricular aspects and are conducted every year.
Most of the schools plan, prepare, and maintain activity calendar for the students.
Moral values, social ethics, tolerance and acceptance of all the communities of the
society, and similar values should be nurtured in the schoolchildren. This could not be
imbibed in them just by giving theoretical lectures. It could be inculcated only
through activity-based learning. It teaches the importance of discipline, culture and
social relationships to the students
Participation in such activities help students to develop and raise their self-confidence.
Hence, these activities are essential part of the curriculum. In the initial pandemic-
phase, not much was thought about these activities but when we had to face recurrent
phases of the lockdown then it became necessary to think on the conduction of these
activities. All these activities were then conducted online.
245
The questions were framed in the survey to know if the students were interested in
participating in these activities, as they were prior the pandemic. In addition, it was
crucial to know that those students who were participating were able to enjoy those
activities or not.
From the survey results, it was noticed that the students were reluctant to participate
in online co-curricular activities. In addition, they did not find those activities
interesting when conducted online. 821 (54.66%) students showed negative impact
on the co-curricular activity participation. 621(41.34%) students gave neutral
response. Very less that is only 60 (3.99%) students showed positive preferences
towards the participation in these activities.
Fig 6.26: Jotform Result for Effect on Co-Curricular Activities Analysis

Analyzing the impact on the extracurricular activities of the students
Since ages, extracurricular activities play a vital role in shaping the personality of the
students. The talent that an individual possess could flourish through the platform that
is provided to the children at school, college, and university level. Inter-school, Inter-
College, inter-university, district, state, national level competitions are conducted for
students.
246
There are children who are good at performing art such as music, dance, drama,
instruments etc. Some are well versed in drawing, painting, and sculpture. While
some are good at sports. Especially at college level, these students get exposure to
various competitions, which might turn to be an opportunity for them to achieve a
major milestone of their career in sports or arts.
Schoolchildren are made aware to all the artforms, indoor, outdoor games so that they
could find their area of interest. Even if not every student have a potential to become
an artist or sportsperson, this exposure helps him or her to find a hobby for
themselves. Participation in extracurricular activities brings a positive change in the
personality of an individual. Research have proved that participation in such activities
symbolizes wellbeing of a human being.
During the pandemic, all these activities got a setback. Certain activities were
conducted online as the lockdown prolonged but the children were not satisfied with
it. In contrast to this, there were students who suddenly developed interest in certain
extracurricular activities and started learning it.
The questions in the survey tried to capture the opinion of the students regarding their
interest in online activity participation. In addition, their responses were collected to
know if they enjoyed participating in these activities.
The graph plot for extracurricular analysis showed that majority of the students
responded neutrally to the question. 910 (60.59%) students were in the category of
neutral effect. This could be interpreted that they had not the interest in their hobby or
extracurricular activity but somewhere they were not able to enjoy those activities to
the fullest in online platform. 413 (27.50%) students reflected a negative expression
towards the question. This showed that they were not ready to accept and participate
online. Then the third class of student were those who were very much positive and
enjoyed participation in these activities regardless of the mode.
179 (11.92%) students reflected positive attitude towards the extracurricular activity
participation.
247
Fig 6.27: Jotform Result for Effect on Extracurricular Activities Analysis
Estimating the average emotion of the student
The survey was conducted to understand the student responses to the above
parameters. At the end of the survey, the photo of the student was taken and then the
form was submitted. Thus, the excel data contained student response along with the
respective image of the student. Through the image attempt was made to understand
the emotions of the student when they filled the form. This attempt was made to
increase the reliability of the study.
The CNN model could predict all the seven basic emotions. The set of emotions were
broadly categorized into positive, negative and neutral emotion. The “Happy”
emotion was interpreted as positive; the “Neutral” emotion was evaluated as neutral.
The remaining five emotions were considered “Negative” emotions.
On an average, it was found that the students reflected “Neutral” emotion while
responding to the survey. The graph plotted for average emotion shows this. 1164
(77.45%) students expressed “Neutral” emotion. 303(20.16%) showed positive
emotion and very less number that is 36 (2.40%) students represented “Negative”
emotion.
248
Fig 6.28: Jotform Result for Effect on Students’ Emotions
Region-wise Analysis of the students
Sample Selection
The students from different colleges of the Pune district were considered for the
survey. Senior and Junior colleges were part of the survey. The students were from
the Arts, Commerce and Science faculty. Undergraduate and Post-graduate students
belonging to different courses were considered. Towards the beginning of the survey,
the age of the student is accepted. This helps to cluster the students into “Elementary”
(below 12 years), “Junior high school” (13-15years), “Senior high school” (16-18
years), ” Undergraduate” and “Postgraduate” (above 18 years).
To collect the answers from the students, especially in the behavior and motivation
category Likert scale was used. The physical category responses were mostly
numeric. The sample for the study were the schools in PCMC, senior and junior
colleges within the vicinity of Savitribai Phule Pune University. The students from 50
schools within Pimpri-Chinchwad area were surveyed.
The colleges are located in different parts of the Pune district under SPPU. To analyze
the responses; two major categories Pune Urban and Pune Rural were made.
Infrastructure availability and its quality were the two prominent factors for imparting
education in online mode. Apart from this, the mental and physical state of the student
249
was another parameter. The data collected from the students through the jotform link
was analyzed for understanding the effect on the students’ physical habits, motivation
and behavior, co-curricular and extracurricular.
The results were separated region-wise as urban and rural and the effect on the five
categories were studied and are presented in the graphical analysis below.
Effect on behavior of rural students
The students who filled the survey belonged to the colleges in the rural region of Pune
district. Around 500 students belonging to rural colleges took the survey. It was
observed that the students reflected the “Neutral” behavior pattern largely. 292
(55.94%) students showed “Neutral” effect on their behavior. Secondly, 156 (29.89%)
students showed “Negative” effect. 74 (14.18%) students were happy and showed
“Positive” behavioral effect on them. The graph of the following figure shows the
results in detail.
Fig 6.29: Analysis of Effect on Behavioral Analysis of Rural Students

Effect on behavior of urban students
The bar chart given in the figure below reflects the effect on the behavioral pattern of
the urban students. Around 1000 college students of urban region participated in the
survey. Maximum students were analyzed with “Neutral” effect. 585 (59.63%)
250
students had “Neutral” behavior. Students belonging to “Negative” effect were 259
(26.40%). Whereas, only 137 (13.97%) students were positively influenced.
Fig 6.30: Analysis of Effect on Behavioral Analysis of Urban Students

Effect on Motivation of Rural students
Among the 500 student responses of rural area 279 (53.45%) students showed
negative impact on motivation. It could be interpreted that due to the uncertainties that
existed in the pandemic students were finding it difficult to focus on academics. In
addition, the exposure to social media, excessive use of mobile phones and the
platforms diverted the students. “Negative” category showed the highest number of
students. Then, the “Neutral” class showed 229 (43.87%) students to be affected. The
lowest number of students 14 (2.68) were positively motivated. Following figure
explains the effect on motivation
251
Fig 6.31: Analysis of Effect on Motivational Analysis of Rural Students
Effect on Motivation of Urban students
In the city areas, it was seen that the Negative impact on student motivation was
more. From the graphical analysis, it was clear that 515 (52.50%) students were
reflecting “negative” impact on student motivation. 424 (43. 22%) students depicted
“Neutral” effect. Only 42 students (4.28) students were positive and were motivated.
Figure below shows result.
252
Fig 6.32: Analysis of Effect on Motivational Analysis of Urban Students
Effect on Physical attributes of students in rural areas
According to the survey carried out in rural areas, it was found that the students
prominently reflected a negative impact on the physical aspects. Around 198
(37.93%) students were negatively affected. 167(31.99%) students showed neutral
effect on their physical attributes. Close to this, around 157 (30.08%) students were
positively influenced. The bar chart in the following figure shows the observed
results.
253
Fig 6.33: Analysis of Effect on Physical Analysis of Rural Students
Effect on Physical attributes of students in urban areas
In the Urban areas, also the negative impact was high on the physical parameters. 387
(39.45%) students were negatively influenced. 305 (31.09%) students were neutral in
their physical patterns. 289 (29.46%) students responded positively on their physical
parameters. The below figure represents the result analysis through the bar diagram.
254
Fig 6.34: Analysis of Effect on Physical Analysis of Urban Students
Effect on co-curricular activities on rural students
Mostly, students enjoy participating in co-curricular activities. However, in pandemic,
the online mode of conduction made them reluctant and this was reflected through the
survey results. Around 282 students that is 50.02 % of the total number of students
surveyed in rural areas had a negative effect. They were not interested in online
participation in the co-curricular activities. Then, the remaining 223 (42.72%)
students were neutral towards the co-curricular activity participation. Very less
number that is only 17 students (3.26%) showed interest and were positive in
participating online. The results are plotted in a bar chart and are shown in the
following figure.
255
Fig 6.35: Analysis of Effect on Co-curricular Activities Analysis of Rural
Students
Effect on co-curricular activities on urban students
In the urban area also students, it was analyzed that the negative impact was high. 539
(55%) students were negatively impacted. 398 (40.61%) students were neutral
towards the participation. Only 43 students (4.39%) students were positively affected.
The subsequent figure shows the analyzed results.
256
Fig 6.36: Analysis of Effect on Co-curricular Activities Analysis of Urban
Students
Effect on extracurricular activities on rural students
Maximum students in the rural region were neutral towards their involvement in
extracurricular activities through digital mode. 308 (59.00%) students were neutral.
148 (28.35%) students had a negative influence on them and were not interested in
involving in the activities. Only 66 students (12.64%) students had a positive
approach towards the extracurricular activities. The results are shown below in figure.
257
Fig 6.37: Analysis of Effect on Extracurricular Activities Analysis of Rural
Students
Effect on extracurricular activities on urban students
In urban areas, also the students showed neutral effect in more number. 602 (61.43%)
students gave a neutral response. 265 (27.04%) students showed a negative impact on
them and their participation in these activities was negatively affected. Very less 113
(11.53%) students were positive towards the change. The bar chart in the following
figure shows the results.
258
Fig 6.38: Analysis of Effect on Extracurricular Activities Analysis of Urban
Students
Average emotions of students in the rural areas
The average emotion depicted for rural area students was most of the times “Neutral”.
Neutral emotion was depicted by maximum that is 405 (77.59%) students. 101
(19.35%) reflected emotions that could be labeled as “Positive” emotions. Only 16
(3.07%) students showed “Negative” emotions.
259
Fig 6.39: Analysis of Effect on Average Emotions of Rural Students
Average emotions of students in the urban areas
In the Urban area students, the results showed more number of students in the
“Neutral” category for average emotions. 759 (77.37%) students had reflected
“Neutral” emotions on an average. 202 (20.59%) students showed the “Positive”
average emotions. Only 20 (2.04%) students were labelled in the “Negative” average
emotion category.
260
Fig 6.40: Analysis of Effect on Average Emotions of Urban Students
6.6 Mapping of Objectives vs Methods Implemented

In the first chapter, we saw that the researcher had started with five objectives in mind
to begin with the research work. As we saw the methods implemented in this chapter,
let us try to map the objectives and corresponding methods implemented.
Objective Methods Implemented
- To address this objective, four ML models were
To study the overall impact of a developed using KNN, SVM, NN and RF
COVID-19 pandemic on algorithms in the Students Survey application.
students of different age groups. - In this application, inputs from students of various
age groups between 10 to 23 years were taken.
- CNN model was developed using FER-2013
To study attitude, temperament, dataset to study the influence on temperament,
motivation, and learning styles. motivation, attitude and learning preferences (online
or offline) of the students.
To analyze deep learning-based - CNN model was used for capturing the facial
algorithms for predictive expressions of the students using continuous video
analysis of sentiments and streaming that was implemented in the Survey
261
behavior of students. application. This was done to fetch the authentic
results from the application and to gain insights
about their behavioral factors.
To analyze and identify face - Jotform survey was designed to collect the data
recognition algorithms to from remotely located students. In this survey, the
capture the real-time image for photo widget was used to capture their facial
the prediction of the behavior of expressions.
students.
- For the three parameters under study (physical,
motivational and behavioral), a self-designed
To study the impact on various questionnaire was formed considering the relevance
parameters in academics, co- of the training dataset.
curricular and extracurricular - For the rest two parameters (co-curricular and
activities. extracurricular), sufficient training dataset could not
be found. Thus, appropriate questions were framed
to get the intended opinion of the students.
Additionally, this research work may be used in real life if we study the overall online
and distance learning (ODL) market in India and try to apply the learnings from this
research work there to enhance the overall experience of ODL. Let us now study
some aspects of ODL market in India in the subsequent section.
6.7 Potential use in online and distance learning (ODL) market:

The awareness and spread of online and distance learning in India has been increasing
both in pre and post pandemic era. However, it can be noted that the post pandemic
growth has been much significant. In the year wise graph below, we can see that the
market size of the online education in India saw a major jump from the year 2019 to
2020.
262
Fig 6.41: Online and Distance Learning (ODL) Market in India [194]
It has seen a growth of almost 3 times to ~ 3 billion USD between years 2019 to 2023.
It is estimated to increase at compounded rate of around 20% per year and by 2027,
the market is anticipated to grow further by around 3.5 billion USD. The intake of
students per year is touching almost 40 million. Understanding the need of the hour,
the government has also increased the recognition of number of higher education
institutions for conducting the ODL from 56 in the year 2021-22 to 79 in the year
2023-24. [197]
Some more analytics that shows how the online learning has grown between the years
2016 to 2021 are given in multiple segments as given ahead. The supplemental
education in primary and secondary schooling has gone up from 73 to 773 million
USD. Coaching to prepare for competitive exams has gone up from 43 to 515 million
USD. Courses for working professionals for their upskilling / reskilling and
certifications have gone up from 93 to 463 million USD. Online higher education
went up from 33 to 184 million USD. Learning alternate language or pursuing a
hobby through online means grew from 5 to 29 million USD. [196]
The online education has gained popularity due to its presence completely on cloud
requiring comparatively lesser cost and effort for accessing it. The SaaS based
solutions provided in education sector are ERP and LMS. While the government has
also launched National Digital Library and National Academic Repository to
encourage online education.
For the working professionals, to grow in the organization continuous learning has
become a norm. Most desired skills for IT professionals nowadays are Data Science,
263
AI, ML, Cloud Computing etc. In some organizations, pursuing long-term courses
like masters in some discipline like computer science can help in promotions.
Given the benefits of the online education platforms, naturally the working
professionals have been affiliated towards it. As per the study, it has been identified
that with changes in technology, the desired skills have also been evolving. Thus, an
estimated 40% workforce needs to be upskilled or reskilled. With the encouragement
from the government in terms of favorable policies and providing platforms like
Diksha, eVidya etc and major thrust from private players, more than 5000 EdTechs
would support the potential upsurge in the eLearning initiatives in India. [195]
Some of the key and popular online education providers in India are Byju’s, Coursera,
Unacademy, Toppr, Intellipaat, Udemy, Khan Academy, Physics Wallah, Simplilearn,
Vedantu, Upgrad, WhiteHat Jr, EDX, along with the higher education institutions
recognized for online and distance learning by the government of India.
It has been observed that though students resumed offline learning, still they continue
to watch educational video lectures and try to get insights from experts in specific
domains. This is quite evident with the rise in the number of subscribers for the online
platforms. Students prefer to view physics videos from the “Physics Wallah”
application. Similarly, for mathematics, tutoring through “Khan Academy” has been
the most suitable one.
The Khan Academy’s revenue has grown from 44 million USD in the year 2016 to
estimated 221 million USD in the year 2023. There have been around 137 million
users by the end of December 2022. Every month, greater than 30 million users go
through the Khan Academy sessions. [198]
Similarly, the Physics Wallah financials have grown almost 10 times as its operating
revenue grew from INR 24.6 Crores in the FY2020-21 to INR 233 Crores in the year
FY2021-22 and now to INR 780 Crores in the FY2022-23. The profits have gone up
from INR 6.92 Crores to INR 98.23 Crores in the same period. They have more than 1
Crore students, over 31300 audio-visual lectures. [199]
The National Education Policy 2020 has permitted 40% online learning option for the
students. This is again a major change, which will support digitization in Indian
education.
The application designed by the researcher tries to evaluate the impact of the online
learning on different parameters of the students’ personality.
264
The academicians, teachers, the vendors of the online education systems need to
check that in this transition period of education system certain important factors like
mental stress borne by the students, excessive exposure to screen time, negative
impact of technology on teenage students, unnecessary browsing of social media,
reduced physical activity, should be carefully addressed.
If one has to implement online education on a serious note then care should be taken
about the time span of the video lectures according to the age group for whom the
lecture or content is to be posted. As it has been understood from the survey that
students do not remain hooked to the online mode for long hours, they start engaging
themselves in other activities. Thus, the factor of focus should be handled.
Secondly, even if the candidate has opted for online course, weekly or monthly
interaction with the teachers should be scheduled. The survey responses that were
collected under the category of motivation showed that the students could not
experience peer pressure nor they could establish any bond with their classmates
when they choose online courses. Hence, such meetings would help them retain the
feeling of oneness and would help them remain socially connected. This would
further reduce their stress and anxiety levels. It also increase their motivation levels.
6.7 Summary
The chapter represents the results of the research work carried out by the researcher.
The outcome of all the machine-learning models, and the analysis of each of the sub-
category is shown. The visualization of the predictions is made. The comparative
analysis of the four machine-learning models is understood through the graphical
representations. The Machine teaching models K-nearest neighbor, Support vector
machine, Neural network, and Random forest, were applied to find the effect of the
pandemic on the students. The five parameters were studied for evaluating the effect.
The results were again classified as “positive”, “negative”, and “neutral”. The output
was also analyzed for the emotion recognition model. This effect was categorized into
three subparts. The overall outcome for the individual is predicted through the
application as well as from the survey method.
The school and college students undertook the survey. The results were generated by
the application and were displayed in the graphical format. The output of the
machine-learning models and the deep learning model was depicted through the
confusion matrix. The researcher tried to map the objectives vs the methods
265
implemented during the research work. At the end, the researcher tried to see how this
research could be leveraged in the real life use cases.
266
CHAPTER 7
SUMMARY AND CONCLUSION
7.1 Introduction
This chapter sketches the complete thesis in a concise manner. It is the last chapter of
the research work and compiles the important outcomes, data interpretations,
limitations, conclusions and recommendations drawn from the study. It also
elaborates how the study could be enhanced further. This is discussed through the
points given in the future scope.
7.2 Summary
We are towards the concluding part of the thesis. This is the moment where the
researcher would like to present a rundown of the research work. The outbreak of
COVID-19 pandemic brought disruptions in the academic world. Children were
exposed to the digital world overnight. The study was carried out to dive deep and
analyze the influence the pandemic made on the children. The student sample that was
taken for study belonged to diverse groups. This diversification was based on the
demography, age, gender, schools, junior and senior colleges. There were many
questions that were raised during the pandemic like to what extent the use of
technology should be allowed for the students?, what kind of digital discipline they
should follow?, how to govern the physical habits of the children. On top of this as the
lockdown extended, the issues related to mental stability of students, teachers and
parents emerged. There were many reasons for this like financial instability and
uncertainty in many sectors.
The upgradation in the learning environment, curriculum curtailment, conduction of
examinations, assessment and evaluation, and several administrative issues had to be
solved or modified by the policy makers during this period. The online education
during the period of pandemic, then the hybrid mode, and lastly bringing the children
back to offline mode saw major challenges, which were uniformly handled by the
teachers, students, and parents.
The world has faced pandemic, epidemics previously also. Isolation is the preliminary
and common practice that is followed in any pandemic situation. However, in 2020
when we faced the pandemic, the advent of technology was an added element, which
made the situation better and different as compared to the previous situations. It
267
opened opportunities for all innovations, researches, and advancements in the field of
technology, machine-learning, data science and many related domains. On the other
hand, we also saw the adverse effects of digital media on children. Thus, it stressed
the need to train the children to make appropriate use of the technology.
Everywhere we could find that the pandemic was affecting the mental and physical
capabilities of the people. Prior to the pandemic, research had shown that the physical
and mental wellbeing is interrelated. They are dependent on the attitude, and aptitude
of the person. In addition, the academic progress of the students depends upon the
overall wellness of the students. This was revealed through the literature review also.
Thus, the researcher studied the physical, motivational, behavioral, co-curricular and
extracurricular activities of the students. There are several aspects that were affected
but these five parameters were taken into consideration.
The study revolves around examining the impact of COVID on the students with
reference to the above factors. The machine-learning and deep learning mechanisms
were applied to explore the impact.
The proposed model has three major modules. In the first module, the machine-
learning approach was implied to develop the model. Random forest, Support vector
machine, K-nearest neighbor, and Neural network were the four algorithms that were
used to individually develop four different modules. Each of these modules further
has three sub-modules. These sub-modules were developed for finding the impact on
the behavior, motivation, and physical habits of the children.
The second module encompasses the emotion recognition part of the model. The
emotions of the students were captured and studied to understand the intensity with
which the students were influenced. The deep learning approach was used for this.
The convolutional neural network was developed in the process.
The third module represented a student survey tool. Both the modules discussed
above were merged into the application. The student survey was framed using a self-
administered questionnaire. The opinion of the students was collected. The questions
were designed to understand the impact on the attributes of the study. The survey was
grouped into the five categories, which were to be studied. The questions were added
in each category. The tool also collected the general information of the student like
name, age, and their overall experience of the COVID-19 pandemic.
The tool generates individual results reflecting the impact of the pandemic on the five
major parameters. At the end, it also analyses the overall impact of COVID on the
268
student. The application results for all the students were bifurcated based on region as
urban and rural. These are represented in graphical format in the sixth chapter.
To collect responses from students residing at remote location a form was designed
with the help of jotform. The collected responses were then analyzed through the tool
and the results were interpreted. Both the application and survey results were
depicted.
The machine-learning algorithms were used to predict the effect on the five factors.
The results were shown through the confusion matrix. The comparative analysis of
the four models was made and the best model was derived. The results of the deep
learning module were interpreted and were explained through the confusion matrix.
The student survey tool analyzes the responses with respect to the age of the students.
Especially the eating and sleeping habits of the children are dependent on the age of
the student. The level of mood swings is more in teenagers. Hence, it was examined
through the tool.
Due to the pandemic and the stressful situations, and negative emotion around, it was
sometimes hard for the children to remain motivated towards education. The
motivation levels were also tested from the tool. The engagement of students in co-
curricular and extracurricular activities contributes to their level of motivation and
academic performance as well as physical fitness. Hence, these factors were analyzed.
In the survey method, the questions were included to find if the students were
interested in participating in these competitions or activities with the same enthusiasm
as before pandemic. It was found in some students that they were positively affected
towards these activities and started learning new technology, or improved their co-
curricular skills. Even some started pursuing their hobbies in better way. Thus, made
full utilization of the time that they got during the pandemic.
There were students who behaved exactly opposite to this or there were some who
remained neutral to the situation. The application tool was designed to check the
effect on each of these parameters individually and then the final effect of the
pandemic on the individual was generated from the application.
7.3 The interpretations of the Machine-learning Model

AI facilitates the development of machines, which learn through experience and
algorithms. The models are first trained through machine-learning algorithms. The
trained model is then saved. In the next step, this model is applied on the test data.
269
The model is then evaluated against the expected results. If there is a lot of difference
between the actual and expected output then the model is too refined.
Sometimes, even other algorithms are also used to train the model and then the model
with maximum accuracy and efficiency is considered and saved. There are two basic
approaches of Machine-learning based on the problem at hand and the type of data.
These approaches are named as Supervised and Unsupervised learning. There are
several machine-learning algorithms. In the research work, four machine leaning
algorithms were used. These were Support vector machine (SVM), K-nearest
neighbors (KNN), Random Forest (RF), Neural Network (NN).
The research study aims to evaluate the changes among students due to the influence
of the pandemic. Out of the several parameters of the student’s personality, five major
parameters were taken up for the study.
7.4 Evaluation of the Physical, Motivational and Behavioral Model

The models were trained, developed and tested using four machine-learning
algorithms. Support vector machine, Random forest, K-nearest neighbor and Neural
network. All these algorithms carried the evaluation of each of the three aspects that is
physical, motivational, and behavioral.
The classification report of the four models have been displayed in the earlier
chapters. The model classifies each of the physical, motivational and behavioral
analysis in to three classes. This is represented by categories in the table. “-1”
represents negative effect on a particular aspect. Likewise, “0” represents neutral, and
“1” represents positive effect.
7.4.1 KNN Model:

 The overall accuracy of the KNN model for behavior prediction was 87%. The
model did not give any predictions in the “positive” category. The accuracy for
the “negative” effect was 83.33% and for the neutral category, it was 86.96%.
 The precision, recall, and F1-score for the “Negative” behavior was 86%, 100%
and 93% respectively. For the “Neutral” effect the 89%, 58%, and 70% were the
values for the precision, recall, and F1-score.
 The overall accuracy for the motivational category was 72% from the KNN
model. Again, for the positive category the model was not able to make
270
predictions. The overall accuracy for the “Negative” category was 70.79%, and
for the “Neutral” category, the accuracy was 73.61%.
 The precision, recall, and F1-score for the “Negative” class was 71%, 82%, and
76%. These scores for the “Neutral” class were 74%, 67%, and, 70% respectively.
Again, for the positive category the model was not able to make predictions.
 The KNN model accuracy for the physical class was 57%. The model accuracy for
the “negative” class was 65.62%. For “neutral” effect, it was 37.50%. Whereas for
“positive” impact it was reflected as 33.33%.
 The precision, recall, and F1-score were 63%, 85%, 72% for the “negative” class.
38%, 31%, 34%, for the “neutral” effect and, 67, 8, and 15 for the “positive” class.
7.4.2 SVM Model:

 The overall accuracy of the SVM model for behavior prediction was 93%. The
accuracy for the negative effect was 94.12%, for the neutral category it was
87.80%. For the positive class it was 100%.
and 96% respectively. The scores were 88%, 84%, and 86% respectively for the
“neutral” category. The model scores in the “positive” category were 100%, 33%,
and 50%.
 The overall accuracy of the SVM model for the motivational influence was 81%.
The accuracy for the “negative” effect was 81.48%, for the neutral category it was
81.25%. For the “positive” class it was not able to give any results.
and 84% respectively. The scores for the “Neutral” effect were 81%, 82%, and
82% respectively. The model was not able to generate any results for the
“positive” effect.
 The overall accuracy of the SVM model for Physical impact prediction was 66%.
The accuracy for the negative effect was 69.23%, for the neutral category it was
50%. For the positive class it was 100%.
and 79% respectively. The scores for the “Neutral” effect were 50%, 39%, and
44% respectively. The model was not able to generate any results for the
“positive” effect were 100%, 25%, and 40%.
271
7.4.3 NN Model:
 The overall accuracy of the NN model for behavior prediction was 94%. The
accuracy for the negative effect was 94.96%, for the neutral category it was
94.74%. It depicted 75% accuracy for the positive class.
and 97% respectively. In the “neutral” class, the scores were 95%, 84%, and 89%
respectively. The model scores in the “positive” category were 75%, 100%, and
86%.
 The NN motivational model gave an overall accuracy of 84%. The accuracy for
the “negative” effect was 86.84%, for the neutral category it was 84.42%. It
reflected 50% accuracy for the positive class.
62%.
 The physical model of NN gave an overall accuracy of 71%. The accuracy for the
“negative” effect was 82.95%, for the “neutral” category it was 51.92%. For
physical category, 66.67% the model gave accuracy.
 The precision, recall, and F1-score for the “Negative” behavior was 83%, for the
entire three sub 83%, and 83% respectively. In the “neutral” class, the scores
were 52%, 55%, and 53% respectively. The model scores in the “positive”
category were 67%, 58%, and 62%.
7.4.4 RF Model
 The RF model gave an overall accuracy of 93% for the behavior prediction. The
accuracy for the” negative” effect was 94.12%, for the “neutral” category it was
87.80%. It depicted 100% accuracy for the “positive” class.
50%.
 The RF model for motivation gave an overall accuracy of 90%. The accuracy for
the” negative” effect was 87.21%, for the “neutral” category it was 93.15%. It
depicted 100% accuracy for the “positive” class.
272
 The precision, recall, and F1-score for the “Negative” behavior were 87%, 97%
57%.
 The RF model for physical effect prediction gave an overall accuracy of 85%. The
accuracy for the” negative” effect was 86.46%, for the “neutral” category it was
82.05%. It showed 84.62% accuracy for the “positive” class.
 The precision, recall, and F1-score for the “Negative” behavior were 86%, 94%
88%.
7.4.5 Deep Learning Module

Deep learning is a sub-category of machine-learning. In case of machine-learning,
human beings explicitly train the machine for a particular task, which they want the
machines to perform for them. This is not needed with deep learning. As deep
learning makes use of ANNs (Artificial Neural Networks). These networks depict the
structure of the human brain. The neurons that are interconnected in the ANNs work
together and process the large amount of data, which is given as input to this network.
There are different artificial networks available.
In the research work, the Convolution neural network (CNN) was used. The deep
learning algorithm was used for the second module. For facial expression analysis, the
CNN was applied.
7.5 Major interpretations of the CNN model

 The CNN model was trained on the kaggle dataset. The model recognized the
seven emotions.
 The precision, recall, and F1-score for all the seven emotions that is angry,
disgust, fear, happy, sad, surprise, and neutral were reflected in the classification
report. It gave 98% for angry and disgust. For the emotion of fear the precision,
recall and F1-score were 95%, 98%, and 97% respectively. For “sad” emotion, it
was 98%, 97%, 98%. For “surprise” 98%, 99%, 99%. For “Neutral” 99%, 98%,
98%. For the “Happy” emotion, it was 99% for all the three values.
273
 The overall accuracy of the CNN model was 98%. The “Happy” emotion was
represented by the “positive” effect. The “Neutral” emotion corresponded to the
neutral effect. The “Negative” emotion was the representation of all the remaining
emotions of fear, anger, disgust, sad, surprise. The model accuracy for “Positive”
emotion was 99% and for “Negative” emotion, it was 99%.
7.6 The Survey application Results

The survey application tool was used to take the input from the students. The
students would fill the survey with the camera switched ON. This helped in
collecting the facial responses of the respondent. The application analyzed the
students on the five parameters mentioned above along with the emotion analysis
through facial expression monitoring. Some interesting results that the researcher got
from the survey tool are jotted below. Around 200 students to fill the survey used the
application tool.
 In the physical category, it was found that 43.35% students had no change in their
physical habits. 33. 53% were adversely affected. In addition, 23.12% were high
on this parameter.
 In terms of motivation, the more number of students reflected stable levels of
motivation or no effect on their motivational level. 88.44% students were in this
class. 10.98% students were having very low motivation level. Hardly, 0.58 %
remained highly motivated.
 The negative impact was high in the behavioral effect. 64.74% students belonged
to this group. 26.01% showed ideal behavioral patterns and 9.25% represented no
change in their attitude.
 56% students were least interested in the co-curricular activities. 29.71% were
very much involved in these activities and wanted to continue it. 14.29% reflected
no major changes.
 In the category of extracurricular activities 61.14% were not interested, 20.57%
were interested and 18.29% had no change in their participation and interests.
 The average emotion was “neutral” for 93.10% students. It was “negative” for
5.75% students, and “positive” for 1.15% students.
7.7 The Jotform Survey Results:
274
 The jotform survey was helpful to collect data of students residing at remote
places. Total responses that were collected were 1,503. The results were analyzed
for all the five categories.
 Out of the 1,503 students who filled the survey, 14.04 % showed a positive
behavior. 58.35% students reflected neutral attitude, and 27.61% showed a
negative effect on the behavior.
 In the motivation category 3.73% students showed positive response. 52.83%
reflected negative impact. 43.45% students showed neutral effect.
 The results of the physical effect were depicted as 29.67 % students were
positively affected in the pandemic. 38.92% students were negatively influenced.
31.40% students showed neutral effect.
 In the co-curricular activity participation only 3.99% students showed positive
attitude. 54.66% students showed negative results. 41.34% students remained
neutral to the change.
 In the category of extracurricular activity, analysis 11.92 % students were positive
towards participation in these activities. 27.50% were not interested in
participation and displayed a negative approach. 60.59% students showed no
effect and were categorized as students having neutral effect.
 Apart from these five categories, the overall effect of the pandemic on the students
was captured. Here, the average emotion of the student was also evaluated.
 The results reflected “Neutral” emotion for 77.45% students. 20.16% students
depicted “Positive” emotion. Only 2.40% students expressed “Negative” emotion.
Thus, the module reflected most of the times “neutral” result.
7.7.1 Region wise Analysis of the jotform data

The students studying in urban as well as rural areas filled the jotform survey. Thus,
the results were further analyzed region wise.
7.7.2 Effect on Rural area students

Around 500 students from the rural areas filled the survey. The results are highlighted
below:
 14.18% students were happy and showed “Positive” behavioral effect on them.
29.89% students showed “Negative” effect. 55.94% students showed “Neutral”
effect on their behavior.
275
 The next category was motivation for which the results were estimated. 2.68%
students were positive about the pandemic and remained motivated. 53.45%
students were negatively affected and were low on motivation. The remaining
students were neutral with respect to the motivational parameter and their
percentage was 43.87%.
 The pandemic imposed a positive impact on the physical parameter of around
30.08% students. 37.93% were negatively affected and 31.99% students showed
neutral effect on their physical attributes.
 For the co-curricular category 50.2%, students were negatively influenced.
42.72% were neutral and 3.26% showed a “positive” impact.
 In the extracurricular activity involvement 59% students were “neutral”, 12.64%
showed positivity and 28.35% were negatively affected.
 The average emotions were monitored for the urban area students. It was found
that the 77.59% of the students had no significant effect on the emotions. 3.07%
were in the category of “negative” emotions. The” positive” emotions were shown
by 19.35% students.
7.7.3 Effect on Urban area students

 In the urban area, 1000 students took up the survey. The impact of the pandemic
on the five attributes is summarized below:
 It was observed that 59.63% remained stable, 26.40% showed negative behavioral
patterns and 13.97% had a positive impact on their attitude.
 The results showed that 52.50% students were low in motivation. 4.28% students
were highly motivated and the rest of the students had no change in their
motivational levels. The percentage of these students was 43.22%.
 Less number of students of the urban areas were involved in physical routine at
the time of pandemic. Their percentage was 39.45%. 31.09% students made no
changes in their physical activities and 29.46% students developed good level of
physical activeness.
 It was found that the co-curricular activity participation by urban area students
was hampered largely. 55% of the students were hardly interested in participating
in online mode. 40.61% students were not affected. In addition, 4.39% students
were ready to participate.
276
 The results of the extracurricular category showed that 27.04% students were least
interested in participation. 61.43% were not affected. 11.53% students were
active.
 The average emotions were observed for the students in the urban areas. It was
found that high number of students that is 77.37% student expresses stable
emotions while taking up the survey. Emotions were taken on average. 20.59%
students showed happy emotion and very few that is 2.04% reflected discontent
and were categorized in the “Negative” average emotion category.
CONCLUSION
To satisfy the objectives set in the research design, the researcher has made some
observations as given below:
7.8 Observations
1. The researcher had observed that there is a need of designing a survey application
tool.
2. This was to fulfill the minimum of the requirements and to get appropriate
outcomes/results.
3. For simplicity, students were categorized into five categories based on their age
groups.
4. Four algorithms were used to evaluate the results. Random Forest, Support Vector
Machine, Neural Network and K-Nearest Neighbor were the algorithms used for
the study.
The model that was developed in the research work generated certain outcomes.
These outputs and major findings of the research would help to setup some
conclusions.
7.9 Conclusions
As per the objectives, hypothesis and observations following conclusions were
drawn
1. To understand the state of mind of the students it is necessary to evaluate the
physical, motivational, and behavioral aspects of the students.
2. As per the results derived, it could be concluded that the Neural network
algorithm gives appropriate results for behavioral changes in students.
3. Random forest algorithm works appropriately on motivational and physical
analysis of the students.
277
4. It was observed that students preferred attending schools and colleges physically
as they could observe, listen, understand, can act upon and give responses and
feedback.
5. Through the survey application, it was found that the maximum number of
students had a “neutral” effect in the physical and motivational category.
6. It was observed in the results of the application that the students had a “negative”
impact on the behavioral patterns.
7. The co-curricular and extracurricular activity factor results also were inclined
more towards the “Negative” impact.
8. The overall average emotions recognized from the tool displayed “Neutral”
impact on more number of students.
9. It was noted that due to lack of interaction, lack of social connect there was
change in the motivation of the students.
10. Students were not willing to participate in co-curricular or extracurricular
activities on online platforms
7.10 Recommendations
Based on the analysis of facial expressions of the students from various age groups,
captured through the ML and DL models and through their responses to the
questionnaire, following are the recommendations by the researcher:
• The attention span of the students during online education is not as high as it is in
the offline mode especially in the younger students. Thus, the duration of each
session need to have an upper limit of half an hour.
• As the attention span is lesser in online mode and students may not comprehend
the content of entire session, there needs to be regular one on one meet between
teacher and student (wherever feasible in ‘in-person’ mode).
• The usage of survey application by the ODL providers would help them to assess
the content and delivery of their sessions. Thus, this can act as a quick and handy
tool to take proactive measure to enhance their sessions and meaningfully engage
the students.
• For the online programs with higher duration (> 4 hours a day), a session on
physical education for half an hour is to be mandated to avoid negative impacts on
physical as well as mental health of the students.
278
• Beyond study, there needs to be some additional engaging / fun oriented sessions
to enhance the interest level of the students.
FUTURE SCOPE
• The research work was carried out on the four categories of students depending on
their age. Junior high school, senior high school, undergraduate and post-graduate
students took up the survey. In future, the application could also be used to
analyze the learning pattern of employees in an organization.
• The study focused only on five parameters of the student personality. The number
of personality dimensions could be increased further.
• The research work was undertaken within the schools and colleges of Pune
district. The geographical area could be extended in the future.
• The responses of students studying in remote colleges were collected by
circulating a jotform link in colleges. The data collected was then given as input to
the survey application. The application then predicted the effect on student with
respect to the five categories considered for prediction. In future, the survey
application could be enhanced so that the data could be collected directly from the
remote location.
• The analysis of the learning patterns was done using machine-learning and deep
learning algorithms. Further, improved algorithms could be integrated to modify
the application.
• The survey was designed using English language. Hence, those students with
basic reading writing skills in English language could participate. The survey
could be modified to accommodate students studying in regional languages.
LIMITATIONS
1. The researcher has restricted the study to four categories of students depending on
their age.
2. Although the responses were quite precise, still as they were collected in online
mode, there may be an element of error.
3. An attempt is made to capture facial expressions while the student is responding
in online mode, still it is a challenge to read the exact psychological state of the
student.
279
4. There are many algorithms to explore and implement for understanding behavioral
patterns but given the scope of the research work, only some of the machine-
learning and deep learning algorithms were implemented.
5. As the survey application was developed from the perspective of research, it may
need to be evolved further for implementing in the real life use cases.
280
REFERENCES
1. https://en.wikipedia.org/wiki/COVID-19_pandemic_in_India
2. https://blog.gale.com/education-during-plagues-and-pandemics-a-historical-
perspective/
3. https://pib.gov.in/PressReleasePage.aspx?PRID=1655697
4. https://pmevidya.education.gov.in/radio.html
5. https://www.microsoft.com/en-in/microsoft-teams/education
6. https://earthwatch.org/stories/teaching-pandemic-how-educators-are-handling-
sudden-shift-distance-
learning?gclid=EAIaIQobChMIg9Ooj6id9wIVcoNLBR2aTQibEAAYBCAAEgKlM
_D_BwE
7. https://www.techtarget.com/searchunifiedcommunications/definition/Cisco-Webex
8. https://www.dummies.com/article/technology/software/other-software/what-is-
gotomeeting-144802/
9. https://www.indiaspend.com/indiaspend-interviews/to-recoup-learning-loss-schools-
should-focus-on-catch-up-not-curriculum-822073
10. https://itpd.ncert.gov.in/pluginfile.php/10327/mod_page/content/42/Integration%20of
%20ICT%20in%20Teaching-Learning%20and%20Evaluation.pdf
11. https://www.thehindu.com/news/national/over-40-govt-schools-dont-have-power-
playgrounds-parliamentary-panel/article61964470.ece
12. https://pmmodiyojana.in/pm-evidya/
13. https://www.orfonline.org/research/regression-in-learning/
14. https://www.brookings.edu/blog/education-plus-development/2021/08/23/how-has-
education-technology-impacted-student-learning-in-india-during-covid-19/
15. https://data.unicef.org/resources/remote-learning-reachability-factsheet/
16. https://swayam.gov.in/
17. https://zoom.us/
18. Nicu Sebe, Michael S.Lew, Ira cohen, Yafei Sun, Theo Gevers, Thomas S. Huang
(2004), “Authentic Facial Expression Analysis”, Sixth IEEE International Conference
on Automatic Face and Gesture Recognition, ISBN: 0-7695-2122-3, Pp 517-522
xxiv
19. Thomas A.Litzinger, Sang HA LEE, JOHN C.WISE, RICHARD M.FELDER (2007),
“A Psychometric Study of the Index of Learning Styles”, Journal of Engineering
Education, Volume 96, Issue 4, ISSN: 2168-9830, Pp.309-319
20. Adel Ben Youssef, MounirDahmani(2008),” The Impact of ICT on Student
Performance in Higher Education: Direct Effects, Indirect Effects, and Organisational
Change”, RUSC.Universities and Knowledge Society Journal, volume 5, num.1, E-
ISSN: 1698-580X, Pp.45-56
21. Mehdi Sagheb Tehrani(2009),” The Results of Online Teaching: A Case Study”,
Information Systems Education Journal, Volume 7, No 42, ISSN: 1545-679X, Pp. 1-9
22. Matthew P. Walker (2009),“ The Role of Sleep in Cognition and Emotion”, Annals
of the New York Academy of Sciences,ISSN 1749-6632, vol 1156, issue 1, Pp 168-
197
23. Sucheta Kolekar, Sriram Sanjeevi, D.S. Bormane (2010), “ Learning Style
Recognition using Artificial Neural Network for Adaptive User Interface in E-
Learning”, IEEE International Conference on Computational Intelligence and
Computing Research, December 28-29,2010, India, ISBN:978-1-4244-5967-4, Pp.1-5
24. Mikko Vesisenaha (2010),” ICT Education and Computer Science Education for
Development-Impact and Contextualization”, IEEE Frontiers in Education
Conference, October 27-30, 2010, USA, ISBN: 978-1-4244-6262-9
25. Renu Nagpal, Pooja Nagpal, Sumeet Kaur (2010), “ Hybrid Technique for Human
face emotion Detection”, International Journal of Advanced Computer Science and
Applications, ISSN 2156-5570, vol 1, issue 6
26. Jagdish Lal Raheja, Umesh Kumar (2010), “Human Facial expression detection from
detected in the captured image using back propagation neural network”, International
journal of computer science and information technology, vol 2, issue 1, Pp 116-123.
27. Plamen Zahariev, Nina Bencheva, GeorgiHristov, YoanaRuseua (2013), “ICT
Convergence Challenges in Education and their Impact on both Instructors and
Students”, IEEE 24th Annual Conference on European Association for Education in
the Electrical and Information Engineering May 30-31,2013, Greece, ISBN: 978-1-
4799-0043-5, Pp.193-197
xxv
28. Wenbin Liang, Tanya Chikritzhs (2013), “ Sleep duration and its Links to
Psychological Distress, Health Status, Physical activity, and Body Mass Index among
a large Representative General Population Sample”, International Journal of Clinical
Medicine, ISSN : 2158284X, vol 4, issue 1, Pp 45-51
29. Shuiwang Ji (2013) Wei Xu, Ming Yang, Kai Yu,“3D Convolutional Neural
Networks for Human Action Recognition”, IEEE Transactions on Pattern Analysis
and Machine Intelligence, Vol 35, issue 1, ISSN: 0162-8828
30. K .H. Wanjale, Amit Bhoomkar, Ajay Kulkarni, Somnath Gosavi (2013), “Use of
Haar Cascade Classifier for Face Tracking System in real Time Video”, International
Journal of Engineering Research & Technology (IJERT), ISSN: 2278-0181, vol 2,
issue 4
31. Geraldine Gray, Colm Mc Guinness, PhilipOwende, Aiden Carthy (2014),” A
Review of Psychometric Data Analysis and Applications in Modelling of Academic
Achievement in Tertiary Education”, The Journal of Learning Analytics, ISSN: 1929-
7750, Pp. 75-106
32. Charles Dziuban, Patsy Moskal, Jessica Thompson, Lauren Kramer, Genevieve
DeCantis, Andrea Hermsdorfer (2015), “Student Satisfaction with Online Learning: is
it a Psychological Contract?” Journal of Asynchronous Learning Network Volume-
19, Issue-2, ISSN: 1939-5256
33. Simin Ghavifekr, Wan Athirah Wan Rosdy (2015), “Teaching and Learning with
Technology: Effectiveness of ICT Integration in Schools”, International Journal of
Research in Education and Science, Volume 1, issue 2, ISSN: 2148-9955, Pp. 175-
191
34. M Varun Kumar, V.G Navaneeth, M Narasimman, L.R Thanveer Ahmed (2015),
“The role of ICT Tools in teaching in India”, Australian Journal of Basic and Applied
Sciences, volume 9, issue 35, ISSN: 1991-8178, Pp. 191-193
35. Shrutika C Dhargave, Apurva Sonak, Vandana Jagtap (2015), in “ Survey paper on
methods of emotion detection from still images”, International Journal of Current
Engineering and Scientific Research (IJCESR), ISSN: 2394-0697, Vol 2, issue 2, Pp
69-77.
xxvi
36. Havan Agrawal, Harshil Mavani (2015), “Student Performance Prediction using
Machine-Learning”, International Journal of Engineering Research & Technology
(IJERT), ISSN: 2278-0181, vol 4, issue 3
37. Anna Sun, Xiufang Chen, “Online Education and its effective Practice: A Research
Review”, Journal of Information Technology Education Research, Volume 15, ISSN:
1539-3585, Pp.157-190
38. Collen Halupa (2016),” RISKS: The Impact of Online Learning and Technology on
Student Physical, Mental, Emotional, and Social Health”, International Conference of
Education, Research and Innovation, ISSN: 2340-1095, Pp.6305-6314
39. Yaman Koseoglu (2016), “ To What Extent Can the Big Five and Learning Styles
Predict Academic Achievement”, Journal of Education and Practice, Volume 7 Issue
30, ISSN: 2222-1735
40. Fatima Zahra Salmam, Abdellah Madani, Mohamed Kissi (2016), “Facial Expression
Recognition using Decision Trees”, 13th International Conference on Computer
Graphics, Imaging and Visualization (CGiV), ISBN: 978-1-5090-0811-7
41. Dr. Poonam Tanwar, Divya Arora, Dhruv Anand (2016), Facial expression detection
using hidden Markov Model, International Journal of Scientific & Engineering
Research, ISSN: 2229-5518, vol 7, issue 12
42. Ratna Astuti Nugrahaeni, Kusprasapta Mutijarsa (2016), “ Comparative Analysis of
Machine-Learning KNN, SVM, and Random Forests Algorithm for Facial expression
classification”, International Seminar on Application for Technology of Information
and Communication, ISBN: 978-1-5090-2326-4
43. Andre Teixeira Lopes, Edilson de Aguiar, Alberto F. De. Souza, Thiago Oliveira-
Santos (2016), “Facial expression recognition with Convolutional Neural Networks
coping with few data and the training sample”, Pattern Recognition, ISSN No: 1873-
5142, Vol 61, Pp 610-628
44. Hidayet Takci, Kali Gurkahraman, Ahmet FiratYelkuvan (2017), “Measurement of
the appropriateness in career selection of the high school students by using data
mining algorithms: A case study”, IEEE Federated Conference on the Computer
Science and Information Systems, ISBN: 978-83-946253-7-5
xxvii
45. Osisanwo F.Y., Akinsola J.E.T, Awodele O, Hinmikaiye J.O, Olakanmi O, Akinjobi J
(2017), “Supervised Machine-Learning Algorithms: Classification and Comparison”,
International Journal of Computer Trends and Technology (IJCTT), vol 48, issue 3,
Pp 128-138
46. Muhammad Ahsan ul Haq, IrumSajjad Dar, Muhammad Aslam, Qaisar Khalid
Mahmood (2017), “Psychometric study of Depression, Anxiety and Stress among
University Students”, Journal of Public Health, ISSN: 2198-1833
47. Viet Anh Nguyen (2017), “The Impact of Online Learning Activities on Student
Learning Outcome in Blended Learning course”, Journal of Information and
Knowledge Management, Volume 16 No. 4, ISSN: 1793-6926
48. Rianne Conijn, Chris Snijders, Ad Kleingeld, Uwe Matzat (2017),“Predicting Student
Performance from LMS Data: A Comparison of 17 Blended Courses Using Moodle
LMS “, IEEE Transactions on Learning Technologies, ISSN No: 1939-1382 ,Vol 10,
issue 1, Pp 17-29
49. Ryan Dale B. Elnar (2018), “Utility Estimates of a needs-based model instrument in
College Students having academic issues: Initial Reports of validity and its
implications to Psychoeducational assessment and intervention”, International Journal
of advanced research, ISSN: 2320-5407, Pp 874-883
50. Amit Kumar Bishnoi (2018), “Applications of Deep Learning and Machine-
Learning”, Journal of Emerging Technologies and Innovative Research, ISSN No:
2349-5162, Vol 5, issue 11, Pp 285-288
51. Charvi Jain, Kshitij Sawant, Mohammed Rehman, Rajesh Kumar (2018),”Emotion
Detection and Characterization using Facial features”, 3rd International Conference
and Workshops on Recent Advances and Innovations in Engineering (ICRAIE),
ISBN: 978-1-5386-4525-3
52. Zhanjia Zhang, Weiyun (2018), “A Systematic Review of the Relationship between
Physical Activity and Happiness”, Journal of happiness studies, ISSN No: 1389-4978,
Pp 1305-1322
53. Rekha Nagar, Dr. Yudhvir Singh (2019), “ A literature survey on Machine-Learning
Algorithms”, Journal of Emerging Technologies and Innovative Research, ISSN:
2349-5162, Vol 6, Issue-4, Pp 471-474
xxviii
54. Kirsti Grotan, Erik R Sund, Ottar Bjerkeset (2019),” Mental Health, Academic Self-
Efficacy and study Progress Among College students- The SHOT Study, Norway”,
Frontiers in Psychology, Volume 10, issue 45, ISSN: 1664-1078, Pp.1-11
55. Susmita Ray (2019), “ A Quick Review of Machine-Learning Algorithms”,
International Conference on Machine-Learning, Big Data, Cloud and Parallel
Computing (COMITCon), ISBN: 978-1-7281-0211-5
56. Anjali Jagwani (2019), “A Review of Machine-Learning in education”, Journal of
Emerging Technologies and Innovative Research (JETIR), ISSN: 2349-5162, vol 6,
issue 5, Pp 384-386
57. Wisal Hashim Abdulsalam, Rafah Shihab Alhamdani, Mohammed Najm Abdullah
(2019), “Facial Emotion Recognition from videos using deep convolutional neural
networks”, International Journal of Machine-Learning and Computing, ISSN No:
2010-3700, Vol 9, issue 1
58. Darshana Anil Chopade, Shweta Panjabrao Deshmukh, Mayuri Sanjay Khune,
Chaitanyakumar Mahadeo Manwar, Prof Niraj N Kasliwal (2019), “A Survey:
Django framework”, Special Issue on Research and Innovations in Science and
Engineering, ISSN No: 2454-4248, Vol 5, issue 12s, Pp 62-68
59. Renieri Correia, Eiji Adachi (2019), “Detecting Design violations in Django-based
Web applications”, Conference SBCARS, Brazilian Symposium on Software
Components, Architectures, and Reuse, Pp 33-42
60. Dhwani Mehta, Mohammad Faridul Haque Siddiqui, Ahmad Y. Javaid (2019), “
Recognition of Emotion Intensities Using Machine-Learning Algorithms: A
Comparative Study”, Sensors, Vol 19, issue 8
61. Ning Yan, Oliver Tat Sheung Au (2019), in “Online learning behavior analysis based
on machine-learning, Asian Association of open Universities Journal, ISSN No:
2414-6994
62. Timon Elmer, Kieran Mepham, Christoph Stadfeld (2020), “Students’ under
lockdown: Comparisons of students’ social networks and mental health before and
during the COVID-19 crisis in Switzerland”, PLoS one Journal, ISSN: 1932-6203
63. Chrysi Rapanta, Luca Botturi, Peter Goodyear, Lourdes Guardia, Marguerite Koole
(2020), “ Online University teaching during and after the COVID-19 crisis:
xxix
Refocusing teacher presence and learning activity”, ISSN number 2524-4868, Pp
923-945
64. Stephan Gerhard Huber, Christoph Helm (2020), “COVID-19 and schooling:
evaluation, assessment, and accountability in times of crises-reacting quickly to
explore key issues for policy, practice, and research with the school barometer”,
Educational Assessment, evaluation and accountability, ISSN: 1874-8600, Pp. 237-
270
65. Ida Faridah, FebiRatna Sari, Tri Wahyuningsih, FitraPuriOganda, UntungRahardja,
“Effect Digital Learning on Student Motivation during COVID-19”, IEEE
8thInternational Conference on Cyber and IT Service Management, ISBN: 978-1-
7281-7701-4
66. Joko, Agus Budi Santoso, Supari Muslim, Rina Harimurti (2020), “Effectiveness of
Mobile Learning implementation in increasing student competence and preventing
the spread and impact of COVID-19”, IEEE 3rd International Conference on
Vocational and Electrical Engineering, ISBN: 978-1-7281-7434-1
67. Yizhuo Chen (2020), “The influence of COVID-19 depression and hope on subjective
well-being in Chinese adolescents”, IEEE International Conference on Public Health
and Data Science, ISBN: 978-1-7281-8571-2
68. Maria Chierichetti (2020), “Understanding the role that non-academic factors play on
students’ experience during the COVID-19 pandemic”, IEEE FEES World
Engineering Education Forum-Global Engineering Deans Council, ISBN: 978-1-
7281-9756-2
69. Viet Duong, Jiebo Luo, Phu Pham, Tongyu Yang, Yu Wang (2020), in “The Ivory
Tower Lost: How College Students Respond Differently than the General Public to
the COVID-19 Pandemic”, IEEE/ACM International Conference on Advances in
Social Networks Analysis and Mining, ISBN: 978-1-7281-1056-1
70. Ashley L Ruba, Pollak Seth D, “Children’s emotion inferences from masked faces:
Implications for social interactions during COVID-19”, PLoS One, ISSN: 1932-
6203, Vol 15, issue 12
71. NurbaitySabri, NurHamiza Musa, NurNabilah Abu Mangshor, Shafaf Ibrahim,
HazwaHanim Mohamed Hamzah (2020), “ Student Emotion Estimation Based on
xxx
Facial Application in E-Learning during COVID-19 Pandemic”, International Journal
of Advanced Trends in Computer Science and Engineering, ISSN No: 2278:3091,
Vol 9, issue 1.4, Pp 576-582.
72. AdamyaShyam, Nitin Mukesh (2020), “A Django-based educational resource sharing
website: Shreic”, Journal of Scientific Research, ISSN No: 2070-0245, Vol 64, issue
1, Pp 238-252.
73. AnuradhaKhattar, Priti Rai Jain, S.M.K Quadri, “ Effects of the Disastrous Pandemic
COVID-19on Learning Styles, Activities, and Mental Health of Young Indian
Students- A Machine-Learning Approach”, International Conference on Intelligent
Computing and Control Systems (ICICCS), ISBN No: 978-1-7281-4876-2.
74. LijiaChen, Pingping Chen, Zhijian Lin (2020), “Artificial Intelligence in Education:
A Review”, IEEE Access, ISSN No: 2169-3536, Vol 8, Pp 75264-75278.
75. Katerina Kedraka, Christos Kaltsidis (2020), “Effects of the COVID-19 Pandemic on
University Pedagogy: Students’ Experiences and Considerations”, European Journal
of Education Studies, ISSN No: 2501-1111, vol 7, issue 8, Pp 17-29.
76. Andi WahyuIrawan, Dwisona, Mardi Lestari, Psychological Impacts of students on
Online Learning during the Pandemic COVID-19”, KONSELI (E-Journal), ISSN
No: 2355-8539, Vol 7, issue 1, Pp 53-60.
77. Dr. Pravat Kumar Jena (2020), “Impact of Pandemic COVID-19 on Education in
India”, International Journal of current Research, ISSN No: 0975-8333X, Vol 12,
issue 7, Pp 12582-12586.
78. Toshika Pareek, Dr Kiran Soni, (2020) “A Comprehensive Study on COVID-19
Pandemic: An Impact on School Education in India”, Amity Journal of Management,
ISSN No: 2581-5261, Vol 8, issue 2, Pp 49-57.
79. Seema Patil (2020), “ICT plays a vital role in social and educational sector on
accidental occurrence of COVID 19”, International Journal of Scientific Development
and Research (IJSDR), ISSN No: 2455-2631, Vol 5, issue 5, Pp 523-526.
80. Ashish Chandiramani, Pawan Singh (2021), “Management of Django Web
Development in Python”, Journal of Management and Service Science, Vol 1, Issue
2, Pp 1-17.
xxxi
81. F.M. Javed Mehedi Shamrat, Anup Majumder, Probal Roy Antu, Saykot Kumar
Barmon, Itisha Nowrin, Rumesh Ranjan (2021), “Human Face Recognition applying
Haar Cascade Classifier”, International Conference on Pervasive Computing and
Social Networking, Vol 317, Pp 1-15.
82. Rouvrais Siegfried, Gerwel Proches Cecile, Audunsson Haraldur, Chelin Nathalie,
LiemInggriani (2020), “Preparing 5.0 engineering students for an unpredictable Post-
COVIDWorld”, IEEE IFEES World Engineering Education Forum-Global
Engineering Deans Council, ISBN: 978-1-7281-9756-2
83. Peter Kolar, Filip Turcinovic, Dario Bojanjac (2020), “Experiences with online
Education during the COVID-19 Pandemic-Stricken Semester”, IEEE International
Symposium on Electronics in Marine, ISBN: 978-1-7281-5973-7
84. Nabil Hasan Al-Kumain, Abdulsalam Alhazmi, Fathey Mohammed, Nadhim A.
Gazem, Muhammad Salman Shabbir, Yousef Fazea (2021), “Exploring the Impact of
the COVID-19 Pandemic on University Students’ Learning Life: An integrated
Conceptual Motivational Model for Sustainable and Healthy Online Learning”,
Sustainability Journal of, Volume 13, Issue 5, ISSN: 2071-1050
85. Michal Wilczewski, Oleg Gorbaniuk, Paola Giuri (2021), “ The Psychological and
Academic Effects of Studying from the Home and Host Country during the COVID-
19 Pandemic”, Frontiers in Psychology, ISSN:1664-1078
86. Galina Ilieva, Tania Yankova, Stanislava Klisaroa-belcheva, SvetianaL Vanova
(2021), “Effects of COVID-19 Pandemic on University Students’ learning”,
Information Journal, Volume 12, Issue 4, ISSN: 2078-2489
87. Eduardo Puraivan, Marcelo Leon, Jarnishs Beltran, Fabian Riqueime (2021),
“Emotion-based decision support tool for learning processes An application with
undergraduate students during COVID-19 pandemic”, IEEE 16th Iberian Conference
on Information Systems and Technologies, ISBN: 978-1-6654-3022-7
88. Vartika Bhadana, Anand Singh Jalal, Pooja Pathak (2020), “ A Comparative Study of
Machine-learning Models for COVID-19 Prediction in India”, IEEE 4th Conference
on Information and Communication Technology, ISBN: 978-0-7381-2447-6
89. Cheng-Sheng Yu, Shy-Shin Chang, Tzu-Hao Chang, Jenny Lwu, Yu-Jiun Lin,
Hsiung-FeiChien, Ray Jade Chen (2021), “A COVID-19 Pandemic Artificial
xxxii
Intelligence-Based System with Deep Learning Forecasting and Automatic Statistical
Data Acquisition: Development and Implementation Study”, Volume 23, Issue 5,
Journal of Medical Internet Research, ISSN: 1438-8871
90. Uriel Alan Flores-Juarez, Jesus Antonio Alvarez- Cedill, Teodoro Alvarez- Sanchez
(2021), “Two Facial Emotion Detection Based on Naïve Bayesian Classifier”, Journal
of Theoretical and applied Information Technology, ISSN No:1992-8645, Vol 99,
Issue 23, Pp 5888-5897.
91. Sanna Ruhalahti, Teija Lehto, Susanna Saarinen, LeenaKatto, “Identifying Higher
education First- Year Students’ Reported Experiences Studying during the
Pandemic”, European Journal of Education Studies, ISSN No: 2501-1111, vol 8,
issue 8, Pp 1-21.
92. Jianmei Zhang, Liand Zhu, Simin Li Jing Huang, Zhiyu Ye, Quan Wei, Chunping
Du, “Rural-Urban disparities in knowledge, behaviors, and mental health during
COVID-19 pandemic”, Medicine, Vol 100, issue 13.
93. Rana Saeed Al-Maroof, Noha Alnazzawi, Iman A Akour, Kevin Ayoubi, Khadija
Alhumaid, Nafla Mahdi Alahbabi, Maryam Alnnaimi, Sarha Thabit, Raghad Al
Faisal, Ahmad Aburayya, Said A. Salloum (2021) , “The Effectiveness of Online
Platforms after the Pandemic: Will Face-to-Face Classes affect Students’,
Informatics, Vol 8, issue 4, Pp 1-21.
94. Galina Timokhina, Nikolay Perepelkin, Olga Popova, Natalia Ivashova, Taira
Murtuzalieva, Anatoly Shishkin (2021), “Digital Transformation of Educational
services in COVID-19 Pandemic: Social and Technological Aspects”, 37th IBIMA
Conference, ISBN: 978-0-9998551-6-4.
95. Iman Akour, Muhammad Alshurideh, Barween Al Kurdi, Amel Al Ali, Said Salloum
(2021), “Using Machine-Learning Algorithms to predict people’s intention to use
Mobile Learning Platforms During the COVID-19 Pandemic: Machine-Learning
Approach, JMIR Med Educ, Vol 7, issue 1.
96. Tom Crick, Cathryn Knight, Richard Watermeyer, Janet Goodall (2021) “The
International Impact of COVID-19 and “Emergency Remote Teaching” on Computer
Science Education Practitioners”, IEEE Global Engineering Education Conference
(EDUCON), ISBN No: 978-1-7281-8478-4.
xxxiii
97. Jing Chang (2021), “Online Instruction and Offline Classroom teaching: a study on
Parallel Education Systems”, 9th International Conference on Information and
Education Technology (ICIET), ISBN No: 978-1-6654-1933-8.
98. El-Sayed Atlam, Ashraf Ewis, M.M. Abd El Raouf, Osama Ghoneim, Ibrahim Gad
(2021), “A new approach in identifying the psychological impact of COVID-19 on
University students’ academic performance”, Alexandria Engineering Journal, Vol
61, issue 7, Pp 5223-5233.
99. Ankan Bhattacharyya, Somnath Chaterjee, Shibaprasad Sen, Aleksandr Sinitca,
Dmitrii Kaplun, Ram Sarkar (2021) , “ A deep learning model for classifying human
facial expressions from infrared thermal images”, Scientific Reports, ISSN No:
2045:2322, Vol 11, Issue 1.
100. Arijit Nandi , Fatos Xhafa, Laia Subirats, Santi Forts (2021), “Real-time Emotion
Classification using EEG data stream in E-learning Contexts”, MPDI Sensors, ISSN
No: 1424-8220, Vol 21, Issue 5.
101. Mohammad Nehal Hasnine, Huyen T.T. Bui, Thuy Thi Thu Tran, Ho Tran
Nguyen, Gokhan, Akcapinar , Hiroshi Ueda (2021), “Students’ emotion extraction
and visualization for engagement detection in Online Learning”, Procedia Computer
Science, Vol 192, Pp 3423-3431.
102. Taoufik Ben Abdallah , Islam Elleuch, Radhouane Guermazi (2021), “Student
Behavior Recognition in Classroom using Deep Transfer Learning with VGG-16”,
Procedia Computer Science , Vol 192, Pp 951-960.
103. Bo Yang, Jianming Wo, Gen Hattori (2021), “Face Mask aware Robust Facial
Expression Recognition during the COVID-19 Pandemic”, International Conference
on Image Processing, 978-1-6654-4115-5, Pp 240-244.
104. Akshi Kumar (2021), “Machine-learning for psychological disorder prediction in
Indians during COVID-19 nationwide lockdown”, Intelligent Decision Technologies,
ISSN No: 18724981, Vol 15, No 1, Pp 161-172.
105. Rakesh Kumar Singh, Himanshu Gore, Ashutosh Singh, Arnav Pratap Singh
(2021), in “Django Web Development Simple and Fast”, International Journal of
Creative Research Thoughts, ISSN No: 2320:2882, Vol 9, Issue 5, Pp 808-815.
xxxiv
106. Ankit Verma, Chavi Kapoor, Abhishek Sharma, Biswajit Misra (2021), in “Web
application Implementation with Machine-Learning”, International Conference on
Intelligent Engineering and Management, ISBN No: 978-1-6654-1450-0.
107. Xun Zhou, A.K. Qin, Maoguo Gong, Kay Chen Tan (2021), in “A Survey on
evolutionary construction of Deep Neural networks “, IEEE Transactions on
Evolutionary Computation, ISSN No: 1941-0026, Vol 25, Issue 5, Pp 894-912.
108. Pramoda R, Arun P. S., B.S Athul, Bharath B, Naveen Reddy (2021), in “
Emotion Recognition and Drowsiness Detection using Digital Image Processing and
Python”, International Journal of Scientific Research in Science and technology,
ISSN No: 2395-602X, Vol 8, Issue 3, Pp 1037-1043.
109. Nahal Salimi, Bryan Gere, William Talley & Bridget Irioogbe (2021), “College
students mental health challenges: Concerns and Considerations in the COVID-19
Pandemic”, Journal of College Student Psychotherapy, ISSN: 87568225, Vol. 37,
Issue, Pp 39-51
110. Fei Yan, Nan Wu, Abdullah M lliyasu, Kazuhiko Kawamoto, Kaoru Hirota
(2022), “Framework for identifying and visualizing emotional, atmosphere in online
learning environments in the COVID-19 Era”, Applied Intelligence, Pp 9406-9422.
111. Hanan E. Abdelkader, Ahmed G. Gad, Amr A. Abohany, Shaymaa E. Sorour
(2022), “ An efficient data mining technique for assessing satisfaction level with
online learning for higher education students during the COVID-19”, volume 10,
IEEE Access, ISSN:2169-3536, Pp 6286-6303
112. Yijun Zhao, Yi Ding, Yangqian Shen, Samuel Failing, Jacqueline Hwang (2022),
“Different coping patterns among US graduate and undergraduate students during
COVID-19 pandemic: a machine-learning approach”, International Journal of
environmental research and public health, ISSN: 1660-4601
113. Shuang Xue, Michelle R Kaufman, Xing Zhang, Shunan Xia, Chengchen Niu,
Rui Zhou, Wenjian Xu (2022), in “Resilience and Prosocial Behavior Among
Chinese University Students During COVID-19 Mitigation: Testing Mediation and
Moderation Models of Social Support”, Psychology Research and Behavior
Management , Vol 15, Pp 1531-1543.
xxxv
114. Feriha Fatima Khidri, Hina Riaz, Urooj Bhatti, Kamran Ali Shahani, Faiza
Kamran Ali, Sadia Effendi, Keenjhar Rani, Muhammad Nadeem Chohan (2022), “
Physical Activity, Dietary Habits, and Factors Associated with Depression Among
Medical Students of Sindh, Pakistan, During the COVID-19 Pandemic”, Psychology
Research and Behavior Management , Vol 15, Pp 1311-1323.
115. Swadha Gupta, Parteek Kumar, Raj Kumar Tekchandani (2022), “Facial emotion
recognition based real-time learner engagement detection system in online learning
context using deep learning models”, Multimedia tools application, ISSN No:
13807501, 15737721, vol 82, issue 8, Pp.11365-11394.
116. Yuwei Chen, Jianyu He (2022), “Deep learning based emotion detection”, Journal
of Computer and communications, ISSN No: 2327-5227, Pp. 57-71.
117. Dr. M. Neelakantappa , G. Apoorva, D. Santosh Kumar, A. Akshatha, K. Sai
Manoj (2022) “Students live Behavior Monitoring in online classes using artificial
intelligence”, International Research Journal of Modernization in Engineering
Technology and Science, ISSN No:2582-5208, Vol 4, issue 5, Pp. 1920-1927
118. Ashmina Khan, Prof K.N Hande (2022), “A survey on Prediction and Analysis of
students’ academic performance using Machine-Learning Technique”, Journal for
Research in Applied Science and Engineering Technology, ISSN No: 2321-9653, Vol
10, Issue VI.
119. Ismaila Sanusi, Solomon Sunday Oyelere, Henriikka Vartiainen, Jarkko Suhonen
(2022), “A systematic review of teaching and learning machine-learning in K-12
education”, Education and Information Technologies, ISSN No: 1573-7608, Volume
28, Issue 5, Pp: 5967–5997
120. Katerina Kedraka, Christos Kaltsidis, Natassa Raikou, Thanassis Karalis (2022),
“Considerations for University Pedagogy: Distance Learning One Year after the
COVID-19 Pandemic Outbreak”, Journal of Education and Training Studies, ISSN
No: 2324-8068, Volume 10, Issue 3
121. Mrinal Bhardwaj, Ajay Pradhan, Mohammad Kashif Mohiuddin, Swati Kumari,
Shruti Sinha (2022), “E-Learning: Online Learning Platform”, International Journal
of Creative Research Thoughts, ISSN No: 2320-2882, Volume 10, Issue 5 May 2022,
Pp: 816-820
xxxvi
122. Omran Alharbi, Yousef Alshammari, Awatif AlMutairi (2022), “Review theories
of E-learning”, Advances in Social Sciences Research Journal, ISSN No: 2055-0286,
Vol:9, Issue: 2, Pp: 428-434
123. Amjad Rehman Khan (2022), “Facial Emotion recognition using conventional
machine-learning and deep learning methods: Current achievements, analysis, and
remaining challenges”, MDPI, ISSN No: 2078-2489, Vol:13
124. Pablo Dúo-Terrón, Antonio-José Moreno-Guerrero, José-Antonio Marín-Marín
(2022), “ICT motivation in sixth-grade students in pandemic times- The influence of
Gender and Age”, MDPI, ISSN No: 2078-2489, Vol:12, Issue: 3
125. Raj Sharan Shahi, Dr. Vivek Nath Tripathi, Adbhut Pratap Singh, Harsha Devi
(2022), in “A 3-Year Cohort study of Internet use and Parent-Child relationship
Among Senior Secondary School students”, The Online Journal of Journal of
Distance Education and eLearning, ISSN No: 2147-6454 Vol:10, Issue: 3, Pp: 318-
333
126. Lydia T. Liu, Serena Wanga, Tolani Brittonb, and Rediet Abebe (2022),
“Reimagining the machine-learning life cycle to improve educational outcomes of
students”, the Proceedings of the National Academy of Sciences (PNAS), ISSN No:
0027-8424, Vol 120, issue 9
127. Shijing Yuan, Maolin You (2022), in “Effects of Physical Activity on College
Students’ Subjective Well-Being during COVID-19”, Journal of Epidemiology and
Global health, ISSN No: 2210-6006, Vol 12, issue 10
128. Amel Fawzi Zaki (2022), “Investigating the Impact of Extracurricular activities
on the Academic and Social Skills of University Students in Post-COVID-19: A Case
Study”, The Sharjah International Conference on Education in Post COVID-19,
Future trends in education post COVID -19, ISBN No: 978-981-99-1926-0 ,Pp 251-
262.
129. Ahmad Khairulanam Mohamad, Fariza Khalid, Nurul Aina Jamil (2022), “Co-
curriculum activities during the COVID -19 pandemic and their implications towards
soft skills”, International journal of academic research in Business and social
sciences, ISSN No: 2222-6990, Vol 12, issue 6, Pp 39-51.
xxxvii
130. Peijun Jiang, Ehsan Namaziandost, Zeinab Azizi & Mohammad Hasan Razmi
(2022), “Exploring the effects of online learning on EFL learners’ motivation,
anxiety, and attitudes during the COVID-19 pandemic: a focus on Iran”, Current
Psychology, ISSN: 19364733, 10461310, Vol. 42, Pp 2310–2324
131. Samrah Jamshaid, Noor Bahadar, Kamran Jamshed, Misbah Rashid, Muhammad
Imran Afzal, Lv Tian, Muhammad Umar, Xuechao Feng, Imtiaz Khan, Minru Zong
(2023), “Pre and Post-Pandemic (COVID-19) Mental Health of International
Students: Data from a Longitudinal Study”, Psychology Research and Behavior
Management, ISSN No: 1179-1578, vol 16, Pp. 431-446
132. Reyhan Zeynep Pek, Sibel Tariyan Özyer, Tarek Elhage, Tansel ÖZYER, Reda
Alhajj (2023), “The Role of Machine-Learning in Identifying Students at risk and
Minimizing Failure”, IEEE, ISSN No: 2169-3536, vol 11, Pp. 1224 - 1243.
133. Hanjiya Lyu, Arsal Imtiaz, Yufei Zhao, Jiebo Luo (2023), “Human behavior in
the time of COVID-19: Learning from big data”, Frontiers in Big Data, ISSN no:
2624-909X, Vol. 6.
134. Budi Tiara Novitasari, Sukirno, Mimin Nur Aisyah, Ponty S.P. Hutama (2023),
“The influence of technology readiness and learning motivation on students’
performance in online learning during the COVID 19 pandemic: A story of
Indonesian Higher education students”, Journal Ilmiah Global Education, ISSN:
2723-4665, Vol. 4, Issue 1, Pp 96-102
135. Bastian A. Betthauser, Anders M. Bach-Mortensen & Per Engzell (2023), “A
systematic review and meta-analysis of the evidence on learning during the COVID-
19 pandemic”, Nature Human Behavior, ISSN: 2397-3374, Vol. 7, Pp 375–385
136. Abdellah Ibrahim Mohammed Elfeky and Marwa Yasien Helmy Elbyaly (2023),
“Examining the effects of virtual classroom use inside learning management systems
on enhancing student satisfaction”, Annals of Forest Research, ISSN: 18448135,
20652445, Vol. 66, Issue 1, Pp 1980-1990
137. Rahim Sadigov, Elif Yıldırım, Büşra Kocaçınar, Fatma Patlar Akbulut & Cagatay
Catal (2023), “Deep learning based user experience evaluation in distance learning”,
Cluster Computing, ISSN: 1573-7543
xxxviii
138. Alemayehu D. Taye, Liyousew G. Borga, Samuel Greiff, Claus Vögele, Conchita
D’Ambrosio (2023), “A machine-learning approach to predict self-protecting
behaviors during the early wave of the COVID-19 pandemic”, Nature Portfolio,
ISSN: 1476-4687, Vol.13
139. Shaun Liverpool, Mohammed Moinuddin, Supritha Aithal, Michael Owen, Katie
Bracegirdle, Meggie Caravotta, Rachel Walker, Ciaran Murphy, Vicky Karkou
(2023), “Mental health and wellbeing of further and higher education students
returning to face-to-face learning after COVID-19 restrictions”, PLoS ONE, ISSN:
1932-6203, Vol.18
140. Yanyan Hu, Jingzhi Liu, Zhimin Zhao, Cunjian Bi, Hongmin Cao, He Liu,
Guangyan Yang (2023), “Association between sleep quality and psychological
symptoms: A cross-sectional survey of Chinese university students performed during
the COVID-19 pandemic”, Frontiers, ISSN: 1664-1078, Vol. 14
141. Xinchao Li, Flavian Adhiambo Odhiambo and Dickson Kofi Wiredu Ocansey
(2023), “The effect of student’ online learning experience on their satisfaction during
the COVID-19 pandemic: The mediating role of preference”, Frontiers, ISSN: 1664-
1078, Vol. 14
142. Zi‑Yu Huang, Chia‑Chin Chiang, Jian‑Hao Chen, Yi‑Chian Chen, Hsin‑Lung
Chung, Yu‑Ping Cai, Hsiu‑Chuan Hsu (2023), “A study on computer vision for facial
emotion recognition”, Nature Portfolio, ISSN: 1476-4687, Vol.13
143. Sofia, Arun Malik, Mohammad Shabaz and Evans Asenso (2023), “Machine-
learning based model for detecting depression during COVID-19 crisis”, Scientific
African, ISSN: 2468-2276, Vol. 20
144. Ignacio Aguaded, Arantxa Vizcaíno-Verdú, Victoria García-Prieto, Patricia De-
Casas (2023), “The impact of post-pandemic learning loss on education development:
A systematic review” Review of Communication Research, ISSN: 2255-4165, Vol.
11, Pp 172-189
145. Kardoyo, Joko Widodo, Ahmad Nurkhin, Kusumantoro, Khasan Setiaji (2023),
“Hybrid, Online or Face-to-Face Learning? Experiences during COVID-19
Pandemic”, EAI, ISSN: 2593-7650
xxxix
146. Laia Subirats, Aina Palacios Corral, Sof Ia Pérez-Ruiz, Santi Fort, Go Mez-Mon
Ivas Sacha (2023), “Temporal analysis of academic performance in higher education
before, during and after COVID-19 confinement using artificial intelligence”, PLoS
ONE, ISSN: 1932-6203, Vol.18
147. Lindita Bande (2023), “Lessons from Graduation Project During COVID-19
Pandemic for Future Applications, Methods, and Tools”, The Sharjah International
Conference on Education in Post COVID-19, ISSN: 2193-1801, Pp 77-91
148. Saud Altaf, Rimsha Asad, Shafiq Ahmad, Iftikhar Ahmed, Mali Abdollahian,
Mazen Zaindin (2023), “ A Hybrid Framework of Deep Learning Techniques to
Predict Online Performance of Learners during COVID-19 Pandemic”, MDPI, ISSN
No: 2078-2489, Vol: 15, Issue: 15
149. Nawaf Awadh Alreshidi (2023), “How to make Learning Process Smoother for
Students Through Teaching Practices: Action Research”, Book: Future Trends in
Education Post COVID-19, Springer Nature Singapore, ISBN: 9789819919277,
9789819919260, 9789819919277, Pp 151-162
150. Nivin Hasan et al. (2023), “Evaluating the effectiveness of E-Learning in Private
Schools of Amman, Jordan During and After the Corona Pandemic” Book: Future
Trends in Education Post COVID-19, Springer Nature Singapore, ISBN:
9789819919277, 9789819919260, 9789819919277, Pp 173-180
151. Mei Cao, Yongzhen Teng, Na Shao, Yijin Wu(2023), “The relationship between
home-based physical activity and general well-being among Chinese university
students during the COVID-19 pandemic: the mediation effect of self-esteem”,
Humanities and Social Sciences Communications, ISSN No: 2662:9992
152. Bin Gong , Mingchao, Li, Wei Lv (2023), in “Machine-learning and data
analysis-based study on the health issues post-pandemic”, ISSN No: 14327643
153. https://www.netsolutions.com/insights/machine-learning-in-edtech/
154. https://saturncloud.io/blog/what-is-randomstate-in-
sklearnmodelselectiontraintestsplit-example/
155. https://www.simplilearn.com/tutorials/deep-learning-tutorial/rnn
156. https://aspiringyouths.com/advantages-disadvantages/knn-algorithm/
157. https://aspiringyouths.com/advantages-disadvantages/neural-networks/
xl
158. https://www.analyticsvidhya.com/blog/2021/06/understanding-random-forest/
159. https://scikitlearn.org/stable/modules/generated/sklearn.ensemble.RandomForestC
lassifier.html
160. https://www.mygreatlearning.com/blog/random-forest-algorithm/
161. https://www.datasklr.com/select-classification-methods/multi-layer-perceptron
162. https://www.geeksforgeeks.org/support-vector-machine-algorithm/
163. https://learnOpenCV.com/what-is-face-detection-the-ultimate-guide/
164. https://machinelearningmastery.com/face-recognition-using-principal-component-
analysis/
165. https://iq.opengenus.org/techniques-for-face-recognition/
166. https://www.hindawi.com/journals/cin/2022/6453499/
167. https://towardsdatascience.com/softmax-activation-function-how-it-actually-
works-d292d335bd78
168. https://vitalflux.com/keras-categorical-cross-entropy-loss-function/
169. https://machinelearningmastery.com/how-to-accelerate-learning-of-deep-neural-
networks-with-batch-normalization/
170. https://analyticsindiamag.com/guide-to-tensorflow-keras-optimizers/
171. https://towardsdatascience.com/face-detection-with-haar-cascade-727f68dafd08
172. https://keras.io/api/models/model_training_apis/
173. https://pubmed.ncbi.nlm.nih.gov/24892055/#
174. https://iecscience.org/jpapers/46
175. https://vikaspedia.in/education/teachers-corner/bloom-s-taxonomy-of-learning-
domains
176. https://www.tutorialspoint.com/tensorflow/tensorflow_introduction.htm
177. https://www.geeksforgeeks.org/introduction-to-tensorflow/
178. https://machinelearningmastery.com/introduction-python-deep-learning-library-
tensorflow/
179. https://www.analyticsvidhya.com/blog/2021/11/tensorflow-for-beginners-with-
examples-and-python-implementation/
180. https://www.mygreatlearning.com/blog/python-numpy-tutorial/
181. https://www.tutorialspoint.com/scikit_learn/scikit_learn_introduction.htm
xli
182. https://www.simplilearn.com/tutorials/python-tutorial/matplotlib
183. https://www.geeksforgeeks.org/pyplot-in-matplotlib/
184. https://medium.com/@mygreatlearning/everything-you-need-to-know-about-
vgg16-7315defb5918
185. https://javascript.info/coding-style
186. https://www.javatpoint.com/ruby-features
187. https://www.scientecheasy.com/2022/09/python-syntax.html/
188. https://www.jotform.com/
189. https://www.forecast.app/blog/difference-between-accuracy-precision
190. https://docstore.mik.ua/orelly/other/python2/1.21.htm
191. https://www.pythontutorial.net/python-numpy/what-is-numpy/
192. https://levelup.gitconnected.com/introduction-to-scikit-learn-1792dc6937e0
193. https://jovian.com/harsh-singhal1026/python-pandas-data-analysis
194. https://www.technavio.com/report/online-education-market-in-india-market-size-
industry-analysis
195. https://timesofindia.indiatimes.com/blogs/voices/how-virtual-education-has-
forced-a-decade-of-evolution-in-the-sector-in-just-two-years/
196. https://assets.kpmg.com/content/dam/kpmg/in/pdf/2017/05/Online-Education-in-
India-2021.pdf
197. https://deb.ugc.ac.in/
198. https://www.prosperityforamerica.org/khan-academy-statistics/
199. https://startuptalky.com/physics-wallah-success-story/
xlii
APPENDIX – Questionnaire
Questionnaire for collecting Physical features of the student’s personality

1. What is your Age? (in years)
2. How much time did you spend on sleep? (in hours)
3. How much time have you spent on fitness? (in hours)
4. Specify the 'Number of meals per day' you take during a lockdown.
5. What you observed 'Change in your weight'?
6. Did you face any 'Health issues during lockdown'?
List of questions to capture the motivational aspect of the student
1. How much time is spent on TV? (in hours)
2. How much Time is spent on social media? (in hours)
3. How much time is spent on Online Classes? (in hours)
4. How much time is spent on self-study? (in hours)
5. Do you think your time is well utilized?
6. Give your Rating of the Online Class experience.
Questionnaire for collecting the behavioral responses of the students
1. After COVID-19: Continuous exposure to electronic screens in online learning is
tiring and exhausting
2. University learning contributes to strengthening the social personality of students.
3. Staying home for long periods leads to lethargy and laziness.
4. Prolonged use of e-learning tools often leads to boredom, nervousness, and tension.
5. Some students cannot afford to buy all the necessary digital tools, which is
embarrassing and frustrating.
6. I do not recommend continuing with the online learning model because it is socially
and psychologically unhealthy.
7. Measures of lockdown, closures, and quarantine, brought by COVID-19 caused
stress, frustration, and depression.
8. The volume of assignments via e learning led to confusion, frustration, and poor
performance.
9. Face-to-face interaction contributes significantly to boosting students’ academic
achievement.
xliii
10. Taking quizzes and exams online from home was not comfortable and made me
nervous.
11. Do you find yourself more connected with your family, close friends, and relatives?
xliv

Shibani 25721005

Uploaded by

Copyright:

Available Formats

Shibani 25721005

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Shibani 25721005

Uploaded by

Copyright:

Available Formats

TO STUDY THE IMPACT OF THE COVID-19 PANDEMIC ON

EDUCATIONAL PATTERNS AMONGST STUDENTS USING DEEP

UNDER THE GUIDANCE OF

UNDER THE CO-GUIDANCE OF

DEPARTMENT OF COMPUTER SCIENCE

(Signature of the candidate)

To the best of my knowledge and belief, this thesis

To the best of my knowledge and belief, this thesis

I extend my deepest gratitude to my Ph.D. co-supervisor Dr. Ranjit D. Patil, Principal

Education is the cornerstone of all development and advancement, both personally

Sr. No. Title Page No.

1.1 Introduction ...................................................................................................... 1

1.2 The outbreak of COVID-19 .............................................................................. 1

1.3 Impact of previous pandemics on Education .................................................... 2

1.4 Background of the study ................................................................................... 3

1.5 Evolution of Education ..................................................................................... 4

1.6 The Role of ICT in education ........................................................................... 7

1.7 Data Analytics Tools ...................................................................................... 12

1.8 Educational Schemes and government initiatives during COVID-19 ............ 13

1.9 Online Platforms that gained popularity during COVID................................ 15

1.10 Challenges in continuing Education: ............................................................ 19

1.11 Parent’s Role during Pandemic: ................................................................... 23

1.12 Teacher’s role and challenges: ..................................................................... 24

1.13 Challenges faced by students ........................................................................ 25

1.14 Problem at hand ............................................................................................ 30

1.15 Objectives of Research ................................................................................. 31

1.16 Hypothesis .................................................................................................... 31

1.17 Need of Study ............................................................................................... 32

1.18 Scope of Work .............................................................................................. 32

1.19 Thesis Outline ............................................................................................... 33

2.1 Introduction .................................................................................................... 36

3.1 Introduction .................................................................................................... 85

3.2 The technology used in the Model formation ................................................. 86

3.3 The Role of AI in Education........................................................................... 87

3.4 Significance of Machine-learning in education .............................................. 88

3.5 AI and ML approaches in education............................................................... 89

3.6 Need of Artificial Intelligence and Machine-Learning .................................. 93

3.7 Machine-Learning assistance to teachers ....................................................... 99

3.8 Machine-learning in Behavioral Analysis: ..................................................... 99

3.9 Machine-learning in Motivational Analysis ................................................. 101

3.10 Physical Analysis ........................................................................................ 102

3.11 Analyzing the co-curricular and extracurricular factors ............................. 103

3.13 Working of Machine-Learning: .................................................................. 104

3.14 Broad Categories of Machine-learning: ..................................................... 105

3.15 Data Pre-processing for the proposed model .............................................. 108

3.16 Machine-learning algorithms implemented in the proposed model ........... 111

3.17 Summary..................................................................................................... 127

CHAPTER 4 .............................................................................................................. 129

4.1 Introduction .................................................................................................. 129

4.2 Computer Vision and Image Processing ...................................................... 129

4.3 Emotion Recognition .................................................................................... 131

4.4 Teacher’s perspective in understanding emotions of the student ................. 132

4.7 Facial action coding System ......................................................................... 133

4.8 Facial emotion Recognition Structure .......................................................... 133

4.9 Neural Network ............................................................................................ 138

4.10 Convolutional Neural Networks ................................................................. 139

4.11 Transfer Learning ....................................................................................... 142

4.12 Model Creation ........................................................................................... 142

4.13 Haar Cascade Algorithm ............................................................................ 153

4.14 Adaboosting ................................................................................................ 157