International Journal of ISSN: 2669-2333

IJHEP Higher Education Pedagogies Volume 3, Issue 4

Decision Tree Algorithm Use in Predicting Students’ Academic

Performance in Advanced Programming Course

Ismail O. Muraina1*, Edward A. Aiyegbusi1 and Solomon O. Abam2

1
Department of Computer Science, College of Information and Technology Education, Lagos State
University of Education, Lagos, Nigeria
2
Department of Science and Technology Education, Faculty of Education, Lagos State University,
Lagos, Nigeria
niyi2all@[Link]

ABSTRACT

Students’ academic performance or achievement has from time to time been a subject of discourse to
academicians, scholars, researchers, and educational institutions all over the globe. In this regard, schools
are expected to play major and active roles in ensuring that students have good performance at end of their
programs. Academic performance is normally used to classify or predict how students would be ultimately
capable to withstand and face future challenges after graduation. Students’ academic
performance/achievement in any course of study plays a vital role in contributing to and producing
outstanding students who will be future viable leaders. The use of algorithms to classify and predict
students’ academic performance/achievement is not new in machine learning using different techniques
like neural networks, logistic regression, decision trees, and many more. This study classifies and predicts
with the use of a graphical technique called a Decision Tree. The dataset was built from students’
attendance, practical assessment, assignment, ability to complete a free related course on the internet, test
score, and examination grade; a large data set was used to construct and validate the decision tree
algorithm (CHAID) for the first time. CHAID's overall accuracy, sensitivity, and specificity were assessed
using a two-part dataset. The training test and testing set were compared to see how each stage of the
algorithm compares to the other. Results show that the decision tree algorithm makes classification and
prediction visible and clear with the use of graphics to display the results. Hence, the model built produces
96% accuracy.

Keywords: decision tree, academic achievement, classification, prediction, algorithms

Cite this article as: Muraina, I. O., Aiyegbusi, E. A., & Abam, S. O. (2022). Decision Tree Algorithm Use
in Predicting Students’ Academic Performance in Advanced Programming Course. International Journal of
Higher Education Pedagogies, 3(4), 13-23. [Link]

1. Introduction
For many years, researchers have used a range of techniques to categorize and forecast
students' academic performance/achievement at various levels using various metrics, such as
test scores, grade points, teacher rating scales, assignments, and even dropout tests (Ampofo
& Osei-Owusu, 2015). Further, the authors said that academic instruction is the primary aim
of education and schools are in the position to influence students toward learning,
socialization, and vocational preparedness. Therefore, students' academic success is a key
outcome of education, which is why every nation would strive to include the provision of
quality education among its national goals for education. Educational institutions are being
forced to use education as a vehicle for social change in this trend. As a result, the quality of
pupils a school produces determines its success, just as the success of any educational
institution is determined by how well its students perform on both academic and non-
academic tests. Yusuf (2012) argued that performance should be measured not only in terms

© The Author(s). 2022 Open Access. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License,
which permits unrestricted use, distribution, and redistribution in any medium, provided that the original author(s) and source are credited.
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

of test and examination results but also in whether students have acquired survival skills that
allow them to compete with their peers in the labor markets. According to MolokoMphale &
Mhlauli, (2014), education is seen as a promoter of human development and should be at the
center of any society's life and concern. It is a social artifact that embodies aspirations for the
well-being and development of the society it is meant to serve. The major concern nowadays
is about the type of students schools produce. Academic achievement must be given high
priority because it refers to performance outcomes in intellectual domains taught at school,
college, and university; it serves as an indicator of intellectual education. It is the most
important prerequisite for individual and societal prosperity. This makes academic
achievement a vital issue both for students, teachers, and school managers (Spinath, 2012).
Hence, Martín, (2017) asserted that the academic performance of students is not only
associated with an intellectual quotient (IQ) but there are other multiple variables and
dimensions to which a certain predictive value can be attributed to capture cognitive,
psychomotor, and affective domains of the students.
Machine learning algorithms are considered quite powerful in classification problems and
they are the focus of many studies. The prediction capability of these algorithms is very
important and relevant in decision-making (Mienye, Sun & Zenghui, 2019). According to
Sharma, Himani & Kumar, (2016) decision tree algorithm is regarded as one of the most
popular classification techniques. A decision tree is considered a structure that includes a root
node, branches, and leaf (child) nodes. Each internal node denotes a test on an attribute, each
branch denotes the outcome of a test, and each leaf node holds a class label. The topmost
node in the tree is the root node. One of the most well-liked categorization strategies is the
use of decision trees to anticipate outcomes. Consequently, a decision tree algorithm creates a
categorization and predictive model that can handle both numerical and categorical data and
is straightforward to comprehend, interpret, and display graphically. Decision tree analysis
aids in group characteristic identification, examines correlations between independent
variables and the dependent variable, and simply presents this data. The procedure can be
used to determine classification guidelines for upcoming occurrences, for instance, to identify
students who are likely to perform well at the end of his/her program at the institution. Jorda
& Raqueno, (2019) asserted that a decision tree develops classification systems that predict or
classify future observations based on a set of decision rules such algorithms include: The
Classification and Regression Tree (C&R) Tree, Chi-squared Automatic Interaction
Detection (CHAID), C 5.0 and, Quick, Unbiased, Efficient, Statistical Tree (QUEST). Raut
&Nichat, (2017) added that a decision tree is used to sort educational problems by measuring
the students' performance.
This paper is organized as follows: Section I represents the introduction/background of the
study. Section II reviews related literature concerning the factors that affect the performance
of students in learning. Section III represents the materials and methods used. Section IV
represents the results of the findings. Section V represents the discussion of the results gotten
from the previous section. Finally, Section VI shows the Conclusions.

2. Related Literature
It was observed from past studies that students’ academic gain and learning performance
have been affected by various factors including gender, age, teaching methodology, student's
grades, parents’ social economic status, environmental influence, tuition trend, study pattern,
and, the time reserved for study, accommodation (Ali et al., 2013). Numerous academics
have carried out in-depth analyses of the variables influencing student success at various
study levels. Higher education institutions have been interested in students' academic

14
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

performance and graduation rates (Shahzadi & Ahmad, 2011). In the higher education
community, research on the variables that affect university students' academic success is
becoming increasingly popular. Recently, an investigation revealed how variables like
learning styles, gender, and race affect student performance.
Jijo and Abdulazeez (2021) conducted a study on classification using the decision tree
method for machine learning in 2021. The paper's contents, including the
algorithms/approaches employed, datasets, and results obtained, are thoroughly assessed and
presented. To further highlight the authors' topics and determine the most precise classifiers,
all of the methodologies examined were also discussed. As a result, the applications of
various dataset kinds are explored, and the results are examined.
Academic achievement, according to Steinmayr et al. (2015), embodies performance
outcomes that indicate how well a person has achieved particular objectives that were the
focus of activities in instructional environments, specifically in school, college, and
university. Therefore, academic success should be viewed as a complex construct that
includes several learning domains like cognitive, emotional, and psychomotor. There are a
variety of indicators of academic achievement, including extremely basic ones like grades or
results on a test of academic achievement, as well as cumulative ones like educational
degrees and certificates. According to Kapur (2018), factors that affect students' academic
success include their attendance in class, their homework, tests, and exams, as well as their
participation in competitions and other events.
In a study by Jorda and Raqueno (2019), the dataset was split into a training set and a testing
set. Using IBM SPSS Modeler Version 18.0, the training data was utilized to develop and
validate two decision tree algorithms, C5.0 and Chi-squared Automatic Interaction Detection
(CHAID), based on their overall accuracy and ten-fold cross-validation. The best early
warning system for TUPM to identify students who are at risk for academic failure was
therefore CHAID. As a result, the study concludes that the CHAID modeling algorithm
worked best as a predictive model for identifying students who were likely to be retained in
the COE program as well as those who were academically at risk.
To create classifier models, Mustafa (2016) used four distinct classification methods,
including decision tree algorithms, support vector machines, artificial neural networks, and
discriminant analysis. Accuracy, precision, recall, and specificity performance indicators are
used to compare their results over a dataset made up of student replies to an actual course
review questionnaire. All of the classifier models displayed comparable high classification
performances, it was discovered.
Tripti, Dharminder, & Sangeeta (2014) built a performance prediction model based on
students' social integration, academic integration, and a variety of emotional skills using
multiple classification methodologies. On the dataset, the two algorithms used—Random
Tree and J48 (Implementation of C4.5)—performed well.
The enhancement of prediction/classification approaches, which are used to evaluate skill
proficiency based on academic achievement by the breadth of knowledge, was the focus of a
study done by Mayilvaganan & Kalpanadevi (2014). The C4.5 method, AODE, Naive
Bayesian classifier algorithm, Multi-Label K-Nearest Neighbor technique, and decision tree
algorithm were used to identify the best classification accuracy and analyze student
performance using Weka.
Ali et al. (2013) started a study to look into the variables influencing graduate students'
academic performance at the Islamia University of Bahawalpur Rahim Yar Khan Campus.
Students' grade was taken into account as a dependent variable, and the gender, age, faculty

15
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

of study, education father's or guardian's socioeconomic position, residential location, and

medium of instruction were independent variables. The research found that graduate student's
academic performance was highly influenced by their age, their fathers' or guardians'
socioeconomic level, and their daily study time.

3. Materials and Methods

The dataset was built from students’ attendance, practical assessment, assignment, ability to
complete a free related course on the internet, test score, and examination grade; a training
test set and a testing set were created from the dataset. The decision tree algorithm (CHAID)
was developed and validated using training data, and its overall accuracy, sensitivity, and
specificity were assessed using testing data.
This method is broken down into four key steps: data collection, classification, building a
predictive model, and evaluation. In order to define the student's achievement level
dynamically, the weights of the decision tree were concretized. Additionally, the variables of
assessment are determined by the student's attendance, practical assessment, assignment,
ability to complete a free related course online, test score, and examination grade.
Individual responses from 210 advanced programming course students were categorized
using all the aforementioned factors, including attendance, practical assessment, assignment,
ability to finish a complimentary related course online, test score, and examination grade. To
evaluate the student's academic achievement, the DT-CHAID Algorithm was "conformed"
with the experimental process' parameters, which used the decision tree's weights for the
dynamic selection of the exercises. Because they are straightforward to comprehend and
interpret, simple to display graphically, and able to handle both numerical and categorical
data, decision trees have an advantage over alternative prediction models.
The assessment findings were compared to the grades and overall scores of the student's
performance in the evaluation, and this helped to draw important conclusions.

4. Results

4.1. The use of Cross-Validation

Table 1 displays the specifications and results in information of the model. At specifications,
the growing method used was CHAID, the dependent variable was the decision taken,
gender, test scores, attendance, assignment, online free course, practical test, and final
examination formed the independent variables used, validation type was cross-validation, the
maximum tree depth was 3, the minimum cases in parent node was 10, while the minimum
cases in child node were 5. Also, at results section of Table 1, it shows the independent
variable included (Final Examination, Online Free Course, Practical Test, Assignment, and
Gender), the number of nodes was 11, the number of terminal nodes was 6 while the depth of
the results was 3.

16
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

Table 1.
Model Summary for cross Validation
Specifications Growing Method CHAID
Dependent Variable Decision Taken
Independent Variables Gender, Test Scores, Attendance, Assignment, Online
Free Course, Practical Test, Final Examination
Validation Cross Validation
Maximum Tree Depth 3
Minimum Cases in Parent 10
Node
Minimum Cases in Child 5
Node
Results Independent Variables Final Examination, Online Free Course, Practical Test,
Included Assignment, Gender
Number of Nodes 11
Number of Terminal Nodes 6
Depth 3

Figure 1. Decision Tree with CHAID and 10-fold validation

17
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

The number of students and the percentage of students who passed or failed is provided for
each node. The splits happen in priority order. The final test was the most important
consideration, in this case, therefore the 'parent' node, which had all 210 students, split into
two 'child' nodes, one of which contained the students who took the exam and the other of
which contained those who did not. Assignment and Online Free Courses were the next two
important nodes. It was shown that all 105 students who participated in the online free course
finished and got certified; among those that did not finish the online free course, 40 students
passed while 21 failed. Of those that did not finish the online free course but still partook in
the practical test were 14 and they all passed while those who did not finish the online free
course and also missed the practical test were 21. Similarly, students that did not eventually
write the final examination but did the assignment and passed were 7 while those who failed
were 14. Those students that did not do the final examination or assignment failed were 40.
So, 7 were male students that did the assignment but failed while 7 female students’ passed as
well as 7 females failed. Hence, the Practical test and gender formed the terminal nodes for
the classification. The correct classifications were highlighted in grey.

Table 2.
Path Analysis
Terminal Path Classification Number Number
node correct wrong
3 Written → Finish Passed 105 0
6 Not written → Not done Failed 49 0
7 Written → Not Finish → Partake Passed 14 0
8 Written → Not Finish → Miss Failed 21 0
9 Not written → Not done → Male Failed 7 0
10 Not written → Not done → Female Passed 7 7

Table 2 shows terminal nodes, paths, classification, number of correct classifications, and
number of wrong classifications. The path analysis used helps to estimate a system of
equations in which all of the variables are observed. It assumes perfect measurement of the
observed variables; this implies that only the structural relationships between the observed
variables are modeled. Table 2 summarized that 105 students were correctly classified that
wrote the examinations and finished with passed. Likewise, 49 students were correctly
classified that did not write the examination as well as did not do the assignment that
eventually failed; 14 students wrote the final examination, and partook in practical but did not
finish the online test, yet they passed; similarly, 21 students wrote the final examination, not
finished the online test and as well missed the practical class and failed; in the same vein, 7
male students did not write the final examination, they did not do the online test and failed,
while 7 female students did not write the final examination, did not do the online test and
eventually failed while 7 of the female students still managed to pass the final examination.
This showed 100% classification accuracy was obtained from terminal nodes 3, 6, 7, 8, 9, and
10.

Table 3.
Risk
Method Estimate Std. Error
Re-substitution .033 .012
Cross-Validation .048 .015
Growing Method: CHAID; Dependent Variable: Decision Taken

Table 3 explains the information on the proportion of cases misclassified by the proposed
classification. It shows re-submission and cross-validation estimates with standard errors. The

18
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

initial estimate was .048 (Cross-Validation) [Std. error =.015], then after re-submission, the
estimated result gotten was .033 which was considered better than the initial estimate [Std.
error =.012].

Table 4.
Classification
Observed Predicted
Passed Failed Percent Correct
Passed 126 0 100.0%
Failed 7 77 91.7%
Overall Percentage 63.3% 36.7% 96.7%
Growing Method: CHAID; Dependent Variable: Decision Taken

The classification table (Table 4) summarizes the percentages classified correctly. The model
classified 100% of those students who passed correctly with 126 passed, 0 failed, while only
92% of those students who failed with 7 passed, 77 failed. The overall percentage of those
who passed was 63.3% while the overall percentage of those who failed was 36.7%; the
model accuracy level is 96.7%.

4.2. Using the Same Percentage for Both the Training Test and Testing Set
Table 5 displays the specifications and results in information of the model. At specifications,
the growing method used was CHAID, the dependent variable was the decision taken,
gender, test scores, attendance, assignment, online free course, practical test, and final
examination formed the independent variables used, validation type was split sample of 50%
- 50%, the maximum tree depth was 3, the minimum cases in parent node was 10, while the
minimum cases in child node were 5. Also, it shows the independent variable included (Final
Examination, Online Free Course, Practical Test, and Assignment), the number of nodes was
9, the number of terminal nodes was 5 while the depth of the results was 3.

Table 5.
Model Summary
Specifications Growing Method CHAID
Dependent Variable Decision Taken
Independent Variables Gender, Test Scores, Attendance, Assignment,
Online Free Course, Practical Test, Final
Examination
Validation Split Sample [50,50]
Maximum Tree Depth 3
Minimum Cases in Parent 10
Node
Minimum Cases in Child 5
Node
Results Independent Variables Final Examination, Online Free Course, Practical
Included Test, Assignment
Number of Nodes 9
Number of Terminal Nodes 5
Depth 3

19
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

4.3. Information for Training Sample

Figure 2. Decision Tree with CHAID and split sample of (50%) for the training set

20
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

4.4. Information for Testing Sample

Figure 3. Decision Tree with CHAID and split sample of (50%) for the testing set

Table 6.
Risk
Sample Estimate Std. Error
Training .030 .017
Test .036 .018
Growing Method: CHAID; Dependent Variable: Decision Taken

Table 6 explains the information on the proportion of cases misclassified by the proposed
classification. It shows training and testing estimates with standard errors. The Training
estimate was .030 [Std. error =.017], then the Testing estimate result gotten was .036 [Std.
error =.018].

Table 7.
Classification
Sample Observed Predicted
Passed Failed Percent Correct
Training Passed 56 3 94.9%
Failed 0 41 100.0%
Overall Percentage 56.0% 44.0% 97.0%
Test Passed 63 4 94.0%
Failed 0 43 100.0%
Overall Percentage 57.3% 42.7% 96.4%
Growing Method: CHAID; Dependent Variable: Decision Taken

21
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

The classification table (Table 7) summarizes the percentages classified correctly. The model,
at the training set, classified 100% of those students who failed, and 95% of those students
who passed -from which 56 students passed while 3 failed; consequently, the model overall
percentage is 56% passed, 44% failed; 97% overall of accuracy. Likewise, the model at test
set shows 94% of those students passed with 63 students passing, 4 students failed and 100%
of those who failed with 0 passed, 43 failed. So, 57.3% passed, 42.7% failed while the overall
percentage strength is 96% accuracy.

5. Discussion
A decision tree classifier is typically a statistical type of classifier that can be used to cluster
datasets as well as predictions. Nodes and branches in the decision tree help to trace the
classification flow easily. From the analysis, it was observed that the use of different
validation still produced a reasonable result. The results showed that the use of the CHAID
growing method was ideal for the classification. It was also shown that variables like Final
Examination, Online Free Course, Practical Test, and Assignment contributed greatly to the
classification strength of the model while gender contributed less, whereas attendance and
test scores contributed nothing to the model and that is the reason they are excluded in the
model. Hence, the model gave 96% and above accuracy. One of the limitations of this study
is that it is restricted to the use of SPSS to implement the algorithms and covers only one
algorithm out of numerous algorithms available in machine learning.

6. Conclusion
A decision tree is used in the educational field to classify and as well predict students’
academic achievement. It was established in this study that the classification task of a
decision tree is used to evaluate students' performance out of other classification algorithms.
The use of a decision tree enhances the readability and visualization of the results compared
to other classifiers. To predict how well the students would perform on the final exam of the
semester, information about the student's attendance, practical assessment, assignments,
ability to complete a free related course online, test score, and examination grade marks were
directly obtained from the students. This study establishes the reliability of decision trees.
Because they offer classification rules that are simpler to understand than those produced by
other classification techniques, decision trees are quite popular.

References
Ali, S., Haider, Z., Munir, F., Khan, H., & Ahmed, A. (2013). Factors Contributing to the
Student's Academic Performance: A Case Study of Islamia University Sub-Campus.
American Journal of Educational Research, 2013, Vol. 1, No. 8, 283-289, [Link]
10.12691/education-1-8-3
Ampofo, E. T., & Osei-Owusu, B. (2015). Students’ Academic Performance as Mediated by
Students’ Academic Ambition and Effort in the Public Senior High Schools in Ashanti
Mampong Municipality of Ghana International Journal of Academic Research and
Reflection Vol. 3, No. 5, 2015.
Jijo, B. T. & Abdulazeez, A. M. (2021). Classification Based on Decision Tree Algorithm for
Machine Learning; Journal of Applied Science and Technology Trends Vol. 02, No. 01,
pp. 20 – 28 (2021).
Jorda, E. R., & Raqueno, A. R. (2019). Predictive Model for the Academic Performance of
the Engineering Students Using CHAID and C 5.0 Algorithm. International Journal of

22
Muraina, et al., 2022 IJHEP, Vol. 3, No. 4, 13-23

Engineering Research and Technology. ISSN 0974-3154, Volume 12, Number 6 (2019),
pp. 917-928.
Kapur, R. (2018). Factors Influencing the Student’s Academic Performance in Secondary
Schools in India. [Link]
Martín, S. N., Rodrigo, I. G., Izquierdo, G. C., & Ajenjo, P. P. (2017). Exploring Academic
Performance: Looking Beyond Numerical Grades; Universal Journal of Educational
Research 5(7): 1105-1112, 2017. [Link]
Mayilvaganan, M., & Kalpanadevi, D. (2014). Comparison of Classification Techniques for
predicting the performance of Students Academic Environment, International Conference
on Communication and Network Technologies (ICCNT), 2014. [Link]
CNT.2014.7062736
Mienye, I. D., Sun, Y., & Zenghui, W. (2019). Prediction performance of improved decision
tree-based algorithms: a review; 2nd International Conference on Sustainable Materials
Processing and Manufacturing (SMPM 2019). [Link]
011
MolokoMphale, L., & Mhlauli, M. B. (2014). An Investigation on Students' Academic
Performance for Junior Secondary Schools in Botswana. European Journal Of Educational
Research Vol. 3, No. 3, 111-127. [Link]
Mustafa, (2016). Predicting Instructor Performance Using Data Mining Techniques in
HigherEducation, IEEE Access, Volume: 4,2016. [Link]
2568756
Raut, A. B., & Nichat, A. A. (2017). Students Performance Prediction Using Decision Tree
Technique. International Journal of Computational Intelligence Research ISSN 0973-1873
Volume 13, Number 7 (2017), pp. 1735-1741.
Shahzadi, E., & Ahmad, Z. (2011). A Study on Academic Performance of University
Students. Proc. 8th International Conference on Recent Advances in Statistics, Lahore,
Pakistan – February 8-9, 2011, 255-268.
Sharma, H., & Kumar, S. (2016). A Survey on Decision Tree Algorithms of Classification in
Data Mining; International Journal of Science and Research (IJSR) ISSN (Online): 2319-
7064 Index Copernicus Value (2013): 6.14 | Impact Factor (2015): 6.391 Volume 5 Issue
4, April 2016. [Link]
Spinath, B. (2012). Academic Achievement. [Link]
6.00001-X
Steinmayr, R., Meißner, A., Weidinger, A. F., & Wirthwein, L. (2015). Academic
Achievement. Education Oxford Bibliographies; [Link]
6810-0108
Tripti M., Dharminder, K., & Sangeeta, G. (2014). Mining Students’ Data for Performance
Prediction, Fourth International Conference on Advanced Computing & Communication
Technologies,2014. [Link]
Yusuf, A. F. (2012). Influence of principals’ leadership styles on students’ academic
achievement in secondary schools. Journal of Innovative Research in Management and
Humanities, 3(1), 113 -121.

23