Pre-defense

A Comparative Study of Deep Learning Architectures for

Gesture-Based Bangla Hand Sign Recognition

Presented by Supervised by

Sadamun Ahemed Biplob Naznin Sultana

ID: 201-15-14048 Associate Professor
Section: 55-I Department of Computer Science
and Engineering , DIU.

Wednesday 10 January, 2024

 Contents
 Introduction
 Objectives
 Motivations
 Research Questions
 Related Work
 Methodology
 Process Flow Diagram
 Data Collection
 Tokenization Data
 Performance Matrics
 Performance Evaluation
 Limitation
 Future Scope
 Conclusion
 References

Pre-defense
2
 INTRODUCTION
Highlight the risk factors
Define cyberbullying Impact on of cyberbullying and
and harassment on this social media harassment prevalence
modern era of the problem

Classify the text &

characteristics of the any The challenges of
features using ML algorithm addressing cyberbullying
and harassment
And ensemble learning

Pre-defense
3
 OBJECTIVES

 Define the research problem

 Investigated the feasibility of improving cyberbullying detection on a
social network by machine learning
 Provides an in-depth analysis and discussion of our proposed conceptual
prevention framework
 To propose ML based algorithm to prevent cyberbullying from social
media
 Outline the expected outcomes and contributions
 Justify the significance and relevance of the study

Pre-defense
4
 Accurately
detect Global
cyberbully in
real-time problem
tweets.

Motivations
Automatically Communications
identify potential is critical
threats analyzing
the dataset

Tackle this
problem fast,
efficient, and
accurate for the
sake of social
well-being

Pre-defense
5
 RESEARCH QUESTIONS
• Can education be identified as a contributing factor in
reducing instances of cyberbullying?
• How accurately can machine learning models predict the
bullying and not-bullying?
• Which machine learning algorithm performs the best in terms
of accuracy?
• How will this thesis help our country?
• What motivates individuals to engage in cyberbullying
behavior?

Pre-defense
6
 RELATED WORK
Here are some of the insights we found from the related works:
• Most of them used dataset from online repositories.
• Majority of them used the data of twitter based.
• Linguistic features, such as character usage, writing syntax, functional
words
• Comparatively better accuracy was observed using anyone algorithm of
machine learning.
• Some of the models could achieve accuracy around 88%.
• Ensemble model is used very rarely.

Pre-defense
7
 Tools
METHODOLOGY

Define the Developing Discuss

research Collect Data Data pre Analyze the Conclusion of
Data acquisition processing data prediction the study
objectives model

Pre-defense
8
PROCESS FLOW DIAGRAM

Pre-defense
9
 DATA COLLECTION
 For my Cyberbullying and Not-Bullying Prediction task I have collected

8453 data in my own way and also from social media comment .

Pre-defense
10
AFTER TOKENIZATION

Pre-defense
11
PERFORMANCE MATRICS
Naïve SVM
Bayes Decision
Tree

Machine
Ensemble Learning
Learning KNN

Random
Forest SGD

Pre-defense
12
PERFORMANCE MATRICS(Ensemble
Learning)

 Random Forest
 Gradient Boosting
 Voting Classifier
Pre-defense
13
 Performance Evaluation

 Ensemble Learning Accuracy(90%)

Pre-defense
14
Performance Evaluation
(Jaccard Score)

Pre-defense
15
 LIMITATION
Data Privacy Contextual
and Ethics Understanding

Evolution of
Bias and
language and
Fairness
tactics

Pre-defense
16
 FUTURE SCOPE
Improved More Accuracy
Expanded Social Media Platforms
Real Time Monitoring
AI Based Detection Automatically
Increased adaptability to larger datasets
Automatically Restricted For Further
Comment.
Pre-defense
17
 CONCLUSION
 Automatically Classifies Comments.
 Multi-dimensional real training dataset.
 Applying ensemble learning I got better noise free
accuracy.
 Experiment results that SVM performs better.

Pre-defense
18
 REFERENCES
• July 2022 International Journal of Advanced Computer Science and Applications 13(5)
• DOI:10.14569/IJACSA.2022.0130502
• License CC BY 4.0
• T Kanan, A Aldaaja, B Hawashin - Journal of Internet Technology, 2020 - jit.ndhu.edu.tw
• E Raisi, B Huang - Proceedings of the 2017 IEEE/ACM International …, 2017 - dl.acm.org
• YN Silva, C Rich, D Hall - 2016 IEEE/ACM International …, 2016 - ieeexplore.ieee.org
• S Agrawal, A Awekar - Advances in Information Retrieval: 40th European …, 2018 – Springer
• S Mitra, T Tasnim, MAR Islam, NI Khan… - 2021 International …, 2021 - ieeexplore.ieee.org
• I Alanazi, J Alves-Foss - Int Journal of Computer Science and …, 2020 - academia.edu
• DL Hall, YN Silva, B Wheeler, L Cheng… - International Journal of …, 2021 – Springer
• E Raisi, B Huang - … on Advances in Social Networks Analysis …, 2018 - ieeexplore.ieee.org
• YY Liu, P Zavarsky, Y Malik - … , CSS 2019, Guangzhou, China, December 1 …, 2019 – Springer
• D Chatzakou, I Leontiadis, J Blackburn… - ACM Transactions on …, 2019 - dl.acm.org
• S Pericherla, E Ilavarasan - … Software and Networks: Proceedings of INDIA …, 2020 – Springer
• M Ptaszynski, P Dybala, T Matsuba… - International …, 2010 - eprints.lib.hokudai.ac.jp
• J De Angelis, G Perasso - International Journal of Management and …, 2020 - ijmh.org
• KS Alam, S Bhowmik… - 2021 third international …, 2021 - ieeexplore.ieee.org
• S Kim, A Razi, G Stringhini, PJ Wisniewski… - Proceedings of the …, 2021 - dl.acm.org
• EC Ates, E Bostanci, MS Guzel - arXiv preprint arXiv:2101.12718, 2021 - arxiv.org
• S Akhter - 2018 10th International Conference on Electrical …, 2018 - ieeexplore.ieee.org
• THH Aldhyani, MH Al-Adhaileh, SN Alsubari - Electronics, 2022 - mdpi.com
• M Arif - Journal of Information Security and Cybercrimes …, 2021 - journals.nauss.edu.sa

Pre-defense
19
Thank you

Pre-defense
20