International Journal of Distance Education Technologies

Volume 20 • Issue 1

Predicting Student Performance to

Improve Academic Advising Using
the Random Forest Algorithm
Mirna Nachouki, Department of Information Technology, Ajman University, UAE*
https://orcid.org/0000-0003-3490-0878

Mahmoud Abou Naaj, Department of Information Technology, Ajman University, UAE

https://orcid.org/0000-0001-8440-5889

ABSTRACT

The COVID-19 pandemic constrained higher education institutions to switch to online teaching,
which led to major changes in students’ learning behavior, affecting their overall performance. Thus,
students’ academic performance needs to be meticulously monitored to help institutions identify
students at risk of academic failure, preventing them from dropping out of the program or graduating
late. This paper proposes a CGPA predicting model (CPM) that detects poor academic performance
by predicting their graduation cumulative grade point average (CGPA). The proposed model uses
a two-layer process that provides students with an estimated final CGPA, given their progress in
second- and third-year courses. This work allows academic advisors to make suitable remedial
arrangements to improve students’ academic performance. Through extensive simulations on a data
set related to students registered in an undergraduate information technology program gathered over
the years, the authors demonstrate that the CPM attains accurate performance predictions compared
to benchmark methods.

Keywords
Academic Second-and-Third-Year Grades, CGPA Prediction, Educational Data Mining

INTRODUCTION

Student success is one of the main goals of educational institutions. It is measured by academic
performance, or the extent to which students meet the standards defined by these institutions. According
to Tuckman (1975), student performance refers to observing student knowledge, skills, concepts,
understanding, and ideas. In the higher education environment, academic student performance depends
on instructors’ and program coordinators’ standards and reflects the achievement of their short- and
long-term educational goals (Sundar, 2013).
Prediction of accurate academic student performance in universities is considered an important
tool that helps in various decisions related to student admission, retention, graduation, and adapted
educational support based on student data observation. Student performance is a significant indicator
in measuring institutions’ effectiveness and a crucial factor in students’ future success, particularly in

DOI: 10.4018/IJDET.296702 *Corresponding Author

This article published as an Open Access article distributed under the terms of the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/4.0/) which permits unrestricted use, distribution, and production in any medium,
provided the author of the original work and original publication source are properly credited.

1
International Journal of Distance Education Technologies
Volume 20 • Issue 1

countries’ prosperity. For this reason, higher education institutions focus today on improving student
performance and enhancing the quality of their educational programs. An in-depth analysis of the
learners’ previous records can play a vital role in providing quality education to learners.
Early prediction of academic student performance helps institutions provide appropriate actions to
improve students’ retention and success rates. Educational data mining (EDM) involves analysis and
improvement in the prediction methods of student performance. With EDM techniques, researchers
can develop prediction models to detect, monitor, and improve student achievement (Alyahyan &
Düştegör, 2020).
Predicting academic student performance may also improve curriculum content and plan for
adequate academic advising for students. Data mining techniques allow researchers to examine data
sets and obtain conclusions that help improve the educational learning process. Various techniques
have been applied for this purpose. Machine learning, collaborative filtering, Bayesian networks,
artificial neural networks, random forest decision trees, rule-based systems, and correlation analysis
have been applied to predict the risk of dropping out of the university, students’ achievement, or
grades. All these techniques classify the significant factors that affect and foresee overall student
academic performance. However, they differ in precision/accuracy, complexity, and sample data size
requirements.
This paper focuses on developing a prediction model of students’ academic performance based on
their high school average score and second and third-year grades in a four-year information technology
program. It explores the performance of random forest (RF) machine learning in predicting student
performance to achieve high predicting accuracy. The proposed methodology proved its worth by
achieving accurate results. The result can be used by students, advisors, and program coordinators to
reduce education difficulties, improve students’ results, provide better quality education, and develop
plans for education policy.

BACKGROUND
Every year, higher education institutions collect large amounts of student data that could be transformed
into knowledge, which can help instructors, program coordinators, and policymakers analyze and
make adequate decisions. It can also provide timely information to different stakeholders that enhance
the quality of their educational processes. Early student performance prediction can help universities
provide timely actions, like planning for appropriate training to improve students’ learning experience,
thus improving their success rates. In addition, detecting at-risk students early would provide more
time for them to improve their performance (Riestra-González et al., 2021).
Academic performance analysis has gained popularity in the past 20 years. Researchers used
various prediction and classification methods to provide clues to help students improve their
performance and assist educational institutions in improving quality and making better administrative
decisions.

Using Data Mining Techniques to Predict Academic Students’ Success

Data mining (DM) is often used to predict academic students’ performance. It is usually used by

• Program coordinators, to improve curriculum design.

• Academic advisors, to guide their students in choosing the right courses to register for and to
improve their performance in the coming semesters (Han et al., 2010).
• Universities, to improve students’ success rate, reduce retention, and develop future plans for
education policy.

EDM techniques are currently used in various educational situations to measure students’ grades,
understanding, and performance based on their behavior and attitude. EDM uses DM techniques to

2
 International Journal of Distance Education Technologies
Volume 20 • Issue 1

analyze and produce knowledge from an educational data framework (Romero & Ventura, 2007).
EDM is used by different groups of users who can manipulate the extracted knowledge based on their
vision and objectives. For example, instructors can utilize the produced knowledge to improve their
teaching methodologies (Zorrilla et al., 2005). In addition, program coordinators can make better
decisions to enhance learning outcomes (Kotsiantis et al., 2010).
The EDM approach has attracted many researchers to analyze related academic data sets in the
context of high school and university students and then extract relevant information to assist academic
advisors in understanding how students acquire their knowledge and skills and enhance their learning
outcomes (Saa et al., 2019).
Classification is an important DM technique that extracts knowledge from related academic data
sets to classify and predict students’ performance (Chung & Lee, 2019). It consists of discovering
a model categorizing data into classes to predict the class of not yet defined entities. It involves two
phases. In the training phase, a classification algorithm constructs a model by learning from a data
set (training data) and its associated class features—named class label attributes. Then, the previously
built model is applied to classify test data independent of the training data in the test phase.
To predict students’ graduation performance, researchers have applied various classification
algorithms. They concluded that no single classifier technique might be applied in all situations
and provide acceptable prediction results (Asif et al., 2017). Moreover, researchers applied different
strategies, such as decision trees, neural networks, naïve Bayes, k-nearest neighbor, and various others,
to explore the best classification techniques suitable for analyzing specific kinds of data.
Random forest (RF) was considered among the best DM techniques used to predict academic
students’ performance (Rao et al., 2016). This technique has also shown excellent results to predict
high school students’ attrition using several binary classification attributes (Chung & Lee, 2019).
Hussain et al. (2018) presented a similar study using data mining tools and techniques to enhance
students’ academic performance and prevent dropout. Four classification algorithms—J48 decision
tree, Bayes network, and RF—were used. This study used students’ demographic, socioeconomic,
and academic data to perform the prediction. The findings showed that RF outperformed the other
classifiers based on accuracy and classifier errors. Abubakar & Ahmad (2017) predicted students’
performance based on the following prediction features: (1) students’ interaction in an e-learning
environment, (2) their grades, and (3) their prerequisite knowledge. Again, the authors used the RF
algorithm to predict student performance. They concluded that this technique outperformed other
popular data mining techniques, such as decision trees and k-nearest neighbor.
Xu et al. (2017) developed a bilayered structure comprising multiple base predictors and a cascade
of an ensemble predictor for making predictions based on students’ evolving performance states. A
data-driven approach was used in discovering course relevance based on latent factor models and
probabilistic matrix factorization. They showed that the proposed method achieves performance
superior to the benchmark approaches. Furthermore, among the implemented base predictors, they
found that random forest performs the best and k-nearest neighbor performs the worst in most cases.
Hasan et al. (2020) concluded that RF provided a high classification accuracy rate and
outperformed the other algorithms applied (with an accuracy of 88.3%) in predicting successful
students. Rao et al. (2016) compared the performance of the following three algorithms: J48 decision
tree, naïve Bayes, and RF. The findings showed that RF provided better accuracy, especially with
large data set size, though it needed much time to construct the model and provide the results.
Adekitan & Salau (2019) applied a data mining approach to predict students’ final graduation
CGPA using the GPA students got every semester of their first three years of study. They have applied
six different data mining algorithms on the same data set for comparative performance analysis.
According to their results, the logistic regression algorithm produced the best accuracy (89.15%),
followed by tree ensemble (87.884%), decision tree (87.85%), and random forest (87.70%), then naïve
Bayes (86.438%), and finally probabilistic neural network (85.895%). Thus, they have concluded

3
International Journal of Distance Education Technologies
Volume 20 • Issue 1

that engineering students’ final CGPA can be predicted based on their performance in the first six
academic semesters.

Influential Factors in Predicting Students’ Performance

Sixty-nine percent of the studies addressing the prediction of academic performance have cited two
main features: students’ demographics (such as age, gender, race/ethnicity, annual family income,
disability, and parents’ background) and student high school average (Shahiri & Husain 2015;
Almarabeh, 2017; Hamoud et al., 2018; Mengash, 2020; Alturki et al., 2020; Fahd et al., 2021). A
literature review by Alyahyan and Düştegör (2020) revealed that prior academic achievement is the
most crucial factor cited in more than 40% of research. Shahiri and Husain (2015) and Garg (2018)
stated that the high school average is considered the most critical factor that impacts students’
performance reliability. Moreover, this factor influences the prediction of many other different topics
(Aluko et al., 2018). Other factors, such as students’ marks obtained by various course assessment tools
(Almarabeh, 2017; Hamoud et al., 2018), their GPA and CGPA (Ahmad et al., 2015; Hamoud et al.,
2018) as well as their attendance (Almarabeh, 2017) have also been cited in as having a significant
impact on the prediction of academic students’ achievement.
Several researchers applied various machine learning methods to predict student academic
performance at the end of the study program. However, they could not detect which students may
need immediate attention to lower the chances of failing or getting low grades in courses.
The need for this study emerged during the COVID-19 pandemic, when students could not meet
face-to-face with their advisors to discuss issues concerning their academic performance and possible
delay in their graduation. This study contributes to the current body of literature by identifying low-
performing and struggling students by predicting their CGPA at the end of the second year and third
year, and applying various teaching and learning strategies and supplementary measures to improve
their performance.
The high school average score was used in this work as a substitution for demographic and
behavioral factors as many studies have suggested that there is a correlation between high school
performance and non-academic factors (Aluko et al., 2018; Shahiri & Husain 2015; Garg, 2018).

PURPOSE OF THE STUDY

This work aims to investigate whether it is possible to obtain an adequate accuracy in predicting
students’ academic performance based on their high school score average and second- and third-years
grades in a four-year information technology program. The objectives of this study are to:

1. Identify attributes that influence students’ academic performance.

2. Develop a valid model for predicting students’ academic performance.
3. Identify, at early stages, students with low performance.
4. Categorize pillar courses that have a significant impact on students’ academic journey.

SIGNIFICANCE OF THE STUDY

This study is designed to provide academic institutions with a model that will provide decision-makers
with pertinent data to enhance students’ retention and graduation rates. More specifically, it will help:

• Department heads in identifying, at early stages, students at risk of dropping out from the program,
specifically at the end of the second year.
• Advisors in providing dedicated guidance to students with poor academic performance; for
example, recommending students to register elective courses to get them more engaged in the
program.

4
 International Journal of Distance Education Technologies
Volume 20 • Issue 1

• Program coordinators in designing pertinent syllabi for all courses that are considered as pillars
in the program.
• Educators in developing adequate workshops for students to motivate them to improve their
academic performance.

METHODOLOGY
Predicting students’ performance is an essential factor in students’ academic success. However,
students enrolled in the same majors can have different backgrounds, resulting in different sequences
in which they prefer to register for courses. Moreover, not all the courses taken by students are
equally weighted to predict their future performance. For this reason, we will consider IT courses in
the students’ major to decrease the complexity of the problem and reduce the noise that may arise
in the prediction process.
This work examines the use of a random forest algorithm to predict students’ graduation CGPA.
In addition, it uses grades of the second- and third-year courses to measure students’ performance.

Random Forest Algorithm

The random forest algorithm is considered one of the classification learning algorithms that deploy
collections of lower-level classifiers to train and predict autonomous data sets (Gollapudi, 2016). The
algorithm finds a result based on ensemble methods that classify estimators’ results for each predicted
statement. The algorithm finds a result based on ensemble methods that classify estimators’ results
for each predicted statement and uses this result to provide high-level results for distinct fundamental
classifiers. Combining all the results is expected to provide better performance.
A random forest algorithm consists of creating several decision trees. The more decision trees the
algorithm produces, the more accurate the result is. This algorithm creates decision trees on randomly
selected data samples, generates a prediction from each tree, and then uses the voting technique to
select the best solution and the final class of the test data sets (Lee & Chung, 2019). The random forest
algorithm uses a nonlinear technique to discover interrelationships between the various attributes
involved in the problem. The random forest technique has been used to examine several interesting
problems and produce applicable classification models. Applications using this technique are found
in Rodriguez-Galliano et al. (2012). In addition, several researchers have compared he random forest
technique’s performance to other data mining techniques in different problems (Verikas et al., 2011).
Predicting students’ academic performance consists of a typical classification task—the
most commonly applied technique in the educational data mining field (Shahiri & Husain, 2015).
Constructing decision trees for learning and prediction was found to be the most successful technique
in classification-based problems. Random forest or random decision trees are suitable for solving
such problems because a multitude of decision trees will be constructed. Then the selected output
class will be based on the class that they more often predict. Moreover, the RF technique provides
accurate predictions and is widely used in predicting students’ performance in the education sector.
This technique does not prune trees like other tree-based algorithms. Instead, at each tree node, splitting
is considered for a random subset of features, resulting in features being split into more and smaller
random subsets, increasing the diversity among the forest of trees, leading to its outperformance
compared to other decisions trees-based algorithms (Fahd et al., 2021).

Technical Specification
The CGPA predicting model (CPM) is implemented using the Anaconda 3 distribution of Python,
using the scikit-learn and pandas (https://pandas.pydata.org/) libraries, which include a vast array
of functionality for all aspects of data mining and machine learning experimentation and research.
Specifically, the RandomForestClassifier() comprised the core scikit-learn modules used for the
classifier. More specifically, the following Python functions were used:

5
International Journal of Distance Education Technologies
Volume 20 • Issue 1

• describe() computes a summary of statistics relating to the features to identify anomalies.

This function gives the mean, standard deviation, minimum, maximum, and lower and upper
percentiles. The lower percentile is 25, and the upper percentile is 75. The 50th percentile is the
same as the median.
• train_test_split() splits data. As a starting point, we split the data set into training and testing at a
ratio of 70:30. Seventy percent of the data set is applied to train the model, study the relationship
between the features and the target, and fit the model. Secondly, we split the training data into
training and validation at a ratio of 40:30. The purpose of the validation set is to estimate the
prediction error for the selected data. Finally, the test set is used to assess the generalization error
of the model in order to output the corresponding GPA target.
• RandomForestClassifier() instantiates a model with the following parameters: random_state
= 5, n_estimators = 100, criterion: gini, min_sample_split = 2, min_sample_leaf = 1, and
max_features = number of features.
• fit() trains the model on training data.
• predict() uses the model on the test data.

CPM Applicability
The CPM was tested on the educational Kaggle data set (Samin, 2020). This data set contains 525 rows
related to students’ marks in seven different courses (English, botany, zoology, physics, chemistry,
mathematics, and ICT) to predict students’ GPA. First, 70% of the data set (367 records) was applied
to train and validate the model using k-fold (k = 5) cross-validation and study the relationship between
the features (courses) and the target (GPA). The model was then tested by making predictions on a
testing set of 158 records (30%) to output the corresponding GPA target.
It was found that the root mean square error (RMSE) provided by the CPM for GPA predictions
is 0.2723, the mean absolute error is 0.21, and the mean squared error is 7.41, giving 94.29% accuracy
of predicting graduation GPA. These results show that the CPM achieves high prediction on student
performance.

Data Set
Data related to students who have graduated with a four-year bachelor’s degree in information
technology (IT) were used in this study. Data were collected during the academic years 2020–21
and 2021–22 of a private university in the United Arab Emirates (UAE). Only data of students who
have completed the graduation requirements were used to train, validate, and test the model. The
population consists of 105 graduates, of which 73.3% male and 26.7% female, coming from 24
different nationalities, mainly from the Middle East region. All students admitted in the program
have at least a 70% average score in a scientific high school certificate or equivalent. Seventy-four
records were allocated from training using k-fold cross-validation (k = 5), and the remaining were
held out for testing.
The first-year courses consist of general education requirements and introductory IT courses,
which do not reflect students’ actual performance in an IT program. The program’s core courses start
in the second year. Therefore, we have decided to consider students’ grades related to the second- and
third-year core IT courses to predict students’ graduation CGPA in the suggested model, in addition
to their high school average.
Predicting students’ performance at the end of the second year is important to detect, at an early
stage, those students who are at the borderline and need dedicated academic support to succeed in
their educational journey. Third-year prediction is also important as, at this level, students are mature
enough to assume their responsibility and take tangible steps towards enhancing their CGPA a year
before graduation.

6
 International Journal of Distance Education Technologies
Volume 20 • Issue 1

The complete list of features considered in the CPM is detailed in Table 1. Next, the CPM
examines the targets (CGPA) to study the relationships between the features and these targets. Next,
it learns how to predict the CGPA from the features by constructing 100 decision trees. It is then
tested by making predictions on a testing set of data with access to the features (without the targets).
Finally, the predictions are compared to the known answers (targets) to determine how far our CGPA
predictions are from the actual values.

Table 1. Features used in the model

High School Grade Average

Second-year IT courses:
INT201 Object-Oriented Programming
INT202 Discrete Mathematics
INT203 Computer Organization
INT204 Data Structures
INT205 Fundamentals of Data Communications and Networking
INT206 Fundamentals of Web Systems
Third-year IT courses:
INT301 Operating Systems
INT302 Database Management Systems
INT303 Fundamentals of Information Security
INT304 Human–Computer Interaction
INT305 Software Engineering
INT306 Computer Ethics and Professional Development
INT307 IT Project Management
INT308 Enterprise Systems
INT311 Advanced Computer Networks
INT312 Network Security

This process has been executed twice: first with students’ high school average and IT courses
related to the second-year level and a second time with students’ high school average and the second-
and third-year IT courses. We can compare these predictions to the actual value to judge how accurate
the model is.

RESULTS AND ANALYSIS

The CPM was used to predict the CGPA of 30% of the 105 already-graduated students. Table 2
summarizes statistics relating to the mean, standard deviation (std), min, max, and lower (25%) and
upper (75%) percentiles. These numbers demonstrate that no data values seem anomalous, and no
zeros appear in any measurement columns.
We have calculated Pearson correlation and significance one-tailed test of the outcome (CGPA)
and all the variables listed in Table 1. It was found that r (Pearson correlation) is between .351 and
.780. Furthermore, these results show a positive correlation between the outcome (CGPA) and all
variables at a p-value < 0.05 (Table 3).

7
International Journal of Distance Education Technologies
Volume 20 • Issue 1

Table 2. Summary statistics related to the features

50%
Features Mean Std Min 25% 75% Max
(Median)
High School Grade Average 81.62 9.11 56 76 82 88 99
Object-Oriented Programming 79.89 10.87 63 73 77 87 95
Discrete Mathematics 77.24 10.3 63 67 77 87 95
Computer Organization 77.6 11.04 63 67 77 87 95
Data Structures 76.4 10.39 63 67 73 83 95
Fundamentals of Data Communications and
Networking 80.16 9.74 63 73 77 87 95
Fundamentals of Web Systems 82.73 9.19 63 77 83 95 95
Operating Systems 78.33 9.52 63 73 77 83 95
Database Management Systems 80.23 9.39 63 73 77 87 95
Fundamentals of Information Security 81.3 9.92 63 73 83 87 95
Human–Computer Interaction 81.19 8.39 63 77 83 87 95
Software Engineering 78.35 9.75 63 73 77 87 95
Computer Ethics and Professional Development 81.91 8.86 63 73 83 87 95
IT Project Management 78.8 10.23 63 73 77 87 95
Enterprise Systems 83.53 9.93 63 73 83 95 95
Advanced Computer Networks 77.89 10.27 63 73 77 87 95
Network Security 79.72 9.75 63 73 77 87 95

Predicting Students’ Performance Using Second-Year Course Grades

The CPM was first applied to predict the CGPA of graduated students to test its performance. Figure
1 shows the actual and predicted CGPA for each student with regard to their second-year courses
grades. In this figure, the x-axis represents the number of actual and predicted CGPA; the y-axis
represents students’ CGPA ranging between 2.0 and 4.0. In order to obtain these answers, our CPM
uses seven features:

• Students’ performance in high school

• Their grades in
◦◦ Object-Oriented Programming
◦◦ Discrete Mathematics
◦◦ Computer Organization
◦◦ Data Structures
◦◦ Fundamentals of Networking & Data Communication
◦◦ Fundamentals of Web Systems

The RMSE of the CPM’s predictions at the end of the second year is 0.3179, the mean absolute
error is 0.247, and the mean squared error is 10.11, giving 91.32% accuracy of predicting graduation
CGPA.
Figure 2 shows that students’ high school average is the best predictor of their performance at the
end of their second year. The second most important factor is the course Discrete Mathematics. As a
prerequisite to Data Structures, the course Object-Oriented Programming does not highly contribute to

8
 International Journal of Distance Education Technologies
Volume 20 • Issue 1

Table 3. Pearson Correlation and Significance One-Tailed Test of the CGPA and all variables

Pearson Correlation p
CGPA CGPA
CGPA 1.000 .
High School score .351 .000
INT201 .621 .000
INT202 .706 .000
INT203 .752 .000
INT204 .664 .000
INT205 .617 .000
INT206 .622 .000
INT301 .612 .000
INT302 .716 .000
INT303 .673 .000
INT304 .630 .000
INT305 .754 .000
INT306 .592 .000
INT307 .702 .000
INT308 .668 .000
INT311 .739 .000
INT312 .780 .000

Figure 1. Actual and predicted CGPA based on selected features

the prediction of students’ performance, and the accurate level of students’ programming knowledge
and skills is measured after the completion of Data Structures. Finally, this figure shows that the
course Fundamentals of Web Systems is less critical in predicting students’ final CGPA.
In the random forest algorithm, feature importance is calculated by measuring the decrease in node
impurity weighted by the probability of reaching that node. The node probability can be calculated

9
International Journal of Distance Education Technologies
Volume 20 • Issue 1

Figure 2. The predictive power of selected features in ascending order of importance

by the number of samples that reach the node, divided by the total number of samples. The higher
the value, the more important the feature.

Predicting Students’ Performance Using Third-Year Course Grades

To test our CPM’s performance in another situation, we have applied it to predict the CGPA of
graduated students to a broader set of IT core courses. This set of courses are offered in the second and
the third year. Figure 3 shows the actual and predicted CGPA for each student at the end of the third
year. In this current situation, the CPM uses 17 features: students’ performance in high school, and
their grades in the following courses: Object-Oriented Programming, Discrete Mathematics, Computer
Organization, Data Structures, Fundamentals of Data Communication & Networking, Fundamentals
of Web Systems, Database Management Systems, Software Engineering, Operating Systems,
Fundamentals of Information Security, Human–Computer Interaction, IT Project Management,
Enterprise Systems, Computer Ethics & Professional Development, Advanced Computer Networks,
and Network Security.

Figure 3. Actual and predicted CGPA based on all 17 features at the end of the third year

Accuracy and Errors Measures

The RMSE of the CPM’s second run predictions is 0.2984, the mean absolute error is 0.208, and the
mean squared error is 8.9 giving 92.87% accuracy of predicting graduation CGPA.

10
 International Journal of Distance Education Technologies
Volume 20 • Issue 1

Figure 4. The predictive power of all the features in ascending order of importance

Figure 4 shows that the high school grade average maintains the best predictor of students’
performance at the end of their third year.
These two experiments show that the model can indicate students’ final performance early,
allowing the advisor to take appropriate actions early.

Figure 5. Actual and predicted CGPA in both experiments

As per the IT program study plan, Discrete Mathematics, Computer Organization, and Object-
Oriented Programming are offered in the third semester. Data Structures, Fundamentals of Data
Communications, and Fundamentals of Web Systems are offered in the fourth semester. It is worth
mentioning that based on both experiments, the order of importance of these courses in both semesters

11
International Journal of Distance Education Technologies
Volume 20 • Issue 1

is the same (Table 4 (a) and (b); Table 4 (c) and (d)). This finding indicates that the CPM is reliable.
Moreover, in the second experiment, all performance values are improved, as shown in Table 5 and
Figure 5.

Table 4. (a – d) Comparative features importance in the two experiments

Third-Semester Features’ Importance in the First Third-Semester Features’ Importance in the Second
Experiment Experiment
Discrete Mathematics 0.140193 Discrete Mathematics 0.060794
Computer Organization 0.139934 Computer Organization 0.058912
Object-Oriented Programming 0.133045 Object-Oriented Programming 0.055773
(a) (b)
Fourth semester Features Importance in the First Fourth semester Features Importance in the Second
Experiment Experiment
Data Structures 0.135728 Data Structures 0.063488
Fundamentals of Data Communications
0.125054 Fundamentals of Web Systems 0.061614
and Networking
Fundamentals of Data Communications
Fundamentals of Web Systems 0.118749 0.054391
and Networking
(c) (d)

Table 5. Comparative performance of the two experiments

First Experiment Second Experiment

Performance Performance
Root Mean Square Error 0.3179 0.2984
Mean Absolute Error 0.247 0.208
Mean Squared Error 10.11 8.9
Accuracy 91.32% 92.87%

Paired Sample T-Test Between Actual and Predicted CGPA

We conducted a paired sample (dependent) t-test to compare actual students’ CGPA with the predicted
CGPA. Table 6 presents the results of the test.

Hypothesis: No significant difference between students’ actual and predicted CGPA.

The test results showed that actual students’ CGPA and predicted CGPA scores were strongly
positively correlated (r = 0.892). Moreover, the difference was not statistically significant between
actual CGPA (M = 3 .0175, SD = 0.62071) and predicted CGPA scores (M = 2.9668. SD = .65590);
(t31 = .922, p > 0.05). On average, actual CGPA scores were 0.0507 points higher than predicted
CGPA scores (95% Confidence Interval is: -.05910, .15660) Thus, the hypothesis was accepted,
meaning no significant difference exists between students’ actual and predicted CGPA.

12
 International Journal of Distance Education Technologies
Volume 20 • Issue 1

Table 6. Paired sample t-test

DISCUSSION
This proposed CPM found a strong correlation between students’ final CGPA and their high school
average score, with a total importance of 0.207 and 0.076, as shown in Figures 2 and 4, respectively.
Thus, students with higher high school average scores obtain higher final CGPA in their major.
This theory is confirmed by Thiele et al. (2016), who found a strong relationship between students’
contextual background characteristics (such as school grades, school type, and gender) and their
academic performance at university. Therefore, high school average grade provides an adequate
indication of students’ academic performance once they get into university. The findings of this study
can be explained by the fact that, in general, students tend to have the same academic performance
throughout their academic journey. In other words, if a student has relatively good grades in high
school, he/she would continue to have good grades at the university level. Similarly, students having
bad grades in high school are likely to continue getting low performance at university level. Several
research results show a robust relationship between high school grades and college performance and
retention and graduate rates (Asif et al., 2017; Kotsiantis et al., 2010).
The 91.32% and 92.87% accuracy levels obtained in the previous section show that the CPM can
be used as a useful prediction tool to identify students at risk and prevent them from dropping out.
This conclusion is also strengthened by the 94.29% accuracy obtained by the CPM when applied to
the validation data related to the educational Kaggle data set (Samin, 2020).
Adekitan & Salau (2019) have applied the random forest algorithm to predict engineering
students’ final CGPA based on their grades in the program’s first three years with an accuracy of
87.7%. The results obtained by the CPM reflected that students’ final academic performance is

13
International Journal of Distance Education Technologies
Volume 20 • Issue 1

highly related to their high school average. Thus, they represent the strongest predictor of students’
final CGPA consistently throughout the undergraduate four-year information technology program.
Shahiri & Husain (2015) have also applied the decision trees technique to evaluate and predict
students’ performance using course grades, student demographics, and high school background with
an accuracy level of 73%. Amrieh et al. (2016) have applied random forest techniques to predict
students’ performance based on behavioral features. The accuracy level obtained is 75.6%.
The CPM results outperformed the findings obtained by the works mentioned above that use
similar factors in predicting students’ academic performance. Although other studies have considered
various factors in predicting students’ academic performance, the results found using the CPM
are considered satisfactory to provide a significant contribution to department heads, program
coordinators, educators, and advisors to guide their students towards improving their marks.
Furthermore, the CPM results reveal that the second-year courses Discrete Mathematics,
Computer Organization, and Data Structures have a high impact on the students’ final CGPA. This
finding allows academic advisors to give personalized guidance to low-performing students and their
instructors at the end of the first year, such as providing extra course activities to these students and/
or assigning them dedicated assistant instructors to monitor their progress and encourage them to
improve their performance. The same applies to Advanced Computer Networks and Computer Ethics,
which are identified as the most important subjects among the third-year course offerings. These
prompt, appropriate, and personalized actions taken by academic advisors towards low-performing
students would improve students’ final CGPA and prevent them from receiving an academic warning
in their final year of study.
In addition, the CPM results allow the program coordinator to carefully design the content and
assessment tools of the courses that are found to have a major effect on students’ academic performance.

CONCLUSION
This study discusses factors related to students’ academic success or failure in a four-year bachelor of
science program in information technology. The findings are expected to provide insight to university
stakeholders, and more specifically, to program coordinators, student advisors, and instructors, on
the aspects that contribute to students’ academic performance. Understanding the various factors that
impact and delay students’ academic progress is very important; it can help academic institutions design
strategies that can improve students’ academic achievements and enhance the efficiency of education.
This study suggests an educational model, CPM, to predict students’ academic performance,
which various decision-makers could use within higher education institutions. The results generated
indicate that our CPM can be incorporated in an advising tool to identify students that might drop
out of the program. It can also identify students with poor academic performance by predicting their
graduation CGPA based on their grades in core courses in a four-year bachelor’s degree program
and their high school grade average. This prediction provides students estimated final CGPAs, given
their progress using second-year courses grades, which will allow early identification of students
at risk of dropping out from the program. This prediction is then enforced with third-year courses
grades to help academic advisors identify low-performing students and recommend that they register
in elective courses to engage them more in the program. Taking a right corrective step will result in
improving students’ performance, graduation, and retention rates.
The CPM also identifies courses that have an important weight in determining students’ final
CGPA, which helps the program coordinator design the delivery and assessment of these courses
judiciously.

LIMITATIONS
This study revealed the eligibility of the CPM model to efficiently predict students’ performance
at different levels of their studies using their grades in core program courses and their high school
grade average. However, more research is needed to improve the CPM in predicting students with

14
 International Journal of Distance Education Technologies
Volume 20 • Issue 1

low performance by considering various factors affecting students’ performance, specifically those
related to a particular major or a specific institution.
By categorizing and examining these factors, students’ academic performance could be improved
by designing adequate remedial tutorials to enhance the overall educational process. Furthermore,
this study was applied to a small set of data that constitutes the graduates of an IT program; it is
worth mentioning that the CPM was first applied to 82 records and provided an accuracy of 88.68%
and 92.36% in predicting the final CGPA at the end of the second and third years, respectively. This
experiment was then repeated with 105 records; the accuracy of both experiments improved to 91.32%
and 92.87%, respectively. With a larger data set, the RF classification performance would increase.

FUTURE WORK

In the future, the model could be easily and quickly adopted by other higher education institutions, with
larger data sets based on their needs. In addition, other research directions could be explored, such as:

• Implementing different classification algorithms and comparing the results.

• Improving the classification accuracy by using a different hybrid classification algorithm.
• Using multiple data sets from various disciplines to strengthen the validity of the predictive
model.
• Constructing an updated CPM considering demographic and behavioral factors to determine the
impact of these factors on student performance.

Moreover, in this study, we have considered the high school average score only to predict students’
academic performance. In the future, we may expand this research to include the type of high school
certificate (UAE, American, British, French, Indian, etc.).

15
International Journal of Distance Education Technologies
Volume 20 • Issue 1

REFERENCES

Abubakar, Y., & Ahmad, N. B. H. (2017). Prediction of students’ performance in E-learning environment using
Random Forest. International Journal of Innovative Computing, 7(2).
Adekitan, A. I., & Salau, O. (2019). The impact of engineering students’ performance in the first three years on
their graduation result using educational data mining. Heliyon, 5(2), e01250. doi:10.1016/j.heliyon.2019.e01250
PMID:30886917
Ahmad, F., Ismail, N. H., & Aziz, A. A. (2015). The prediction of students’ academic performance using classification
data mining techniques. Applied Mathematical Sciences, 9(129), 6415–6426. doi:10.12988/ams.2015.53289
Almarabeh, H. (2017). Analysis of students’ performance by using different data mining classifiers. International
Journal of Modern Education and Computer Science, 9(8), 9–15. doi:10.5815/ijmecs.2017.08.02
Alturki, S., Hulpuș, I., & Stuckenschmidt, H. (2020). Predicting academic outcomes: A survey from 2007 till 2018.
Technology, Knowledge and Learning, 1-33.
Aluko, R. O., Daniel, E. I., Oshodi, O. S., Aigbavboa, C. O., & Abisuga, A. O. (2018). Towards reliable prediction
of academic performance of architecture students using data mining techniques. Journal of Engineering, Design
and Technology.
Alyahyan, E., & Düştegör, D. (2020). Predicting academic success in higher education: Literature review and
best practices. International Journal of Educational Technology in Higher Education, 17(1), 1–21. doi:10.1186/
s41239-020-0177-7
Amrieh, E. A., Hamtini, T., & Aljarah, I. (2016). Mining educational data to predict student’s cademic performance
using ensemble methods. International Journal of Database Theory and Application, 9(8), 119–136. doi:10.14257/
ijdta.2016.9.8.13
Asif, R., Merceron, A., Ali, S. A., & Haider, N. G. (2017). Analyzing undergraduate students’ performance using
educational data mining. Computers & Education, 113, 177–194. doi:10.1016/j.compedu.2017.05.007
Chung, J. Y., & Lee, S. (2019). Dropout early warning systems for high school students using machine learning.
Children and Youth Services Review, 96, 346–353. doi:10.1016/j.childyouth.2018.11.030
Fahd, K., Miah, S. J., & Ahmed, K. (2021). Predicting student performance in a blended learning environment using
learning management system interaction data. Applied Computing and Informatics. doi:10.1108/ACI-06-2021-0150
Garg, R. (2018). Predicting student performance of different regions of Punjab using classification techniques.
International Journal of Advanced Research in Computer Science, 9(1), 236–241. doi:10.26483/ijarcs.v9i1.5234
Gollapudi, S. (2016). Practical machine learning. Packt Publishing.
Hamoud, A., Hashim, A. S., & Awadh, W. A. (2018). Predicting student performance in higher education institutions
using decision tree analysis. International Journal of Interactive Multimedia and Artificial Intelligence, 5(2), 26–31.
doi:10.9781/ijimai.2018.02.004
Han, J., Kamber, M., & Pei, J. (2011). Data mining: Concepts and techniques (3rd ed.). Morgan Kauffman.
Hasan, R., Palaniappan, S., Mahmood, S., Abbas, A., Sarker, K. U., & Sattar, M. U. (2020). Predicting student
performance in higher educational institutions using video learning analytics and data mining techniques. Applied
Sciences (Basel, Switzerland), 10(11), 3894. doi:10.3390/app10113894
Hussain, S., Dahan, N. A., Ba-Alwib, F. M., & Ribata, N. (2018). Educational data mining and analysis of students’
academic performance using WEKA. Indonesian Journal of Electrical Engineering and Computer Science, 9(2),
447–459. doi:10.11591/ijeecs.v9.i2.pp447-459
Kotsiantis, S., Patriarcheas, K., & Xenos, M. (2010). A combinational incremental ensemble of classifiers as a
technique for predicting students’ performance in distance education. Knowledge-Based Systems, 23(6), 529–535.
doi:10.1016/j.knosys.2010.03.010
Lee, S., & Chung, J. Y. (2019). The machine learning-based dropout early warning system for improving the
performance of dropout prediction. Applied Sciences (Basel, Switzerland), 9(15), 3093. doi:10.3390/app9153093

16
 International Journal of Distance Education Technologies
Volume 20 • Issue 1

Mengash, H. A. (2020). Using data mining techniques to predict student performance to support decision making in
university admission systems. IEEE Access: Practical Innovations, Open Solutions, 8, 55462–55470. doi:10.1109/
ACCESS.2020.2981905
Rao, K. P., Rao, M. C., & Ramesh, B. (2016). Predicting learning behavior of students using classification techniques.
International Journal of Computers and Applications, 139(7), 15–19. doi:10.5120/ijca2016909188
Riestra-González, M., del Puerto Paule-Ruíz, M., & Ortin, F. (2021). Massive LMS log data analysis for the
early prediction of course-agnostic student performance. Computers & Education, 163, 104–108. doi:10.1016/j.
compedu.2020.104108
Rodriguez-Galiano, V. F., Ghimire, B., Rogan, J., Chica-Olmo, M., & Rigol-Sanchez, J. P. (2012). An assessment
of the effectiveness of a random forest classifier for land-cover classification. ISPRS Journal of Photogrammetry
and Remote Sensing, 67, 93–104. doi:10.1016/j.isprsjprs.2011.11.002
Romero, C., & Ventura, S. (2007). Educational data mining: A survey from 1995 to 2005. Expert Systems with
Applications, 33(1), 135–146. doi:10.1016/j.eswa.2006.04.005
Saa, A. A., Al-Emran, M., & Shaalan, K. (2019). Factors affecting students’ performance in higher education: A
systematic review of predictive data mining techniques. Technology. Knowledge and Learning, 24(4), 567–598.
doi:10.1007/s10758-019-09408-7
Samin, T. I. (2020). HSC level students’ GPA dataset. Kaggle. https://www.kaggle.com/tahamidulislamsamin/
final-training
Shahiri, A. M., Husain, W., & Rashid, N. A. (2015). A review on predicting student’s performance using data
mining techniques. Procedia Computer Science, 72, 414–422. doi:10.1016/j.procs.2015.12.157
Sundar, P. P. (2013). A comparative study for predicting students academic performance using Bayesian network
classifiers. IOSR Journal of Engineering.
Thiele, T., Singleton, A., Pope, D., & Stanistreet, D. (2016). Predicting students’ academic performance based on
school and socio-demographic characteristics. Studies in Higher Education, 41(8), 1424–1446. doi:10.1080/030
75079.2014.974528
Tuckman, H. P. (1975). Teacher effectiveness and student performance. The Journal of Economic Education, 7(1),
34–39. doi:10.1080/00220485.1975.10845419
Verikas, A., Gelzinis, A., & Bacauskiene, M. (2011). Mining data with random forests: A survey and results of
new tests. Pattern Recognition, 44(2), 330–349. doi:10.1016/j.patcog.2010.08.011
Xu, J., Moon, K. H., & Van Der Schaar, M. (2017). A machine learning approach for tracking and predicting
student performance in degree programs. IEEE Journal of Selected Topics in Signal Processing, 11(5), 742–753.
doi:10.1109/JSTSP.2017.2692560
Zorrilla, M. E., Menasalvas, E., Marin, D., Mora, E., & Segovia, J. (2005, February). Web usage mining project for
improving web-based learning sites. In International conference on computer aided systems theory (pp. 205–210).
Springer. doi:10.1007/11556985_26

Mirna Nachouki, Ph.D., Member of Association for Computing Machinery (ACM), is the Head of the Information
Technology Department, College of Engineering and Information Technology, at Ajman University. She has obtained
her Ph.D. (1995) and MSc. (1991) in Computer Science from the University of Paul Sabatier, Toulouse III, France
and BSc (1990) in Computer Science from the University of Pau et Pays de l’Adour, France. She served as acting
dean of the former College of Information Technology. She has more than 22 years of teaching and research
experience. Her current research interests include IT Education, Blended Learning, Assessments, Computer
Ethics, Cloud Computing, and Educational Data Mining.

Mahmoud Abou Naaj (Member of IEEE, ACM, and ISI), Ph.D., is an Associate Professor at the Department of
Information Technology, College of Engineering and Information Technology, at Ajman University. He has obtained
his Ph.D. (1983) and MSc. (1979) in Computer Science from the University of Leeds (UK) and BSc (1975) in
Mathematics from the University of Baghdad, Iraq. He served as the College of General Studies Dean, dean of
the College of Information Technology, and Dean of Admission and Registration at Ajman University. He has more
than 35 years of teaching and research experience. His current research interests include IT Education, Blended
Learning, Assessments, Computer Ethics, Computer Algorithms, and Educational Data Mining.

17