DEVELOPMENT OF STUDENT’S ACADEMIC PERFORMANCE PREDICTION

MODEL USING DECISION TREE

Lawal O.O 1,Adepoju A.O2 , Makinde O.E3and Osunade.O. Ph.d4
1,2
Lecturer, Department of Computer Science, D.S Adegbenro I.C.T Polytechnic, Itori , Ogun State
Contact: [email protected], +2348062160129, [email protected], +2348137223966
3
Lecturer, Department of Computer Science, Ajayi Crowther University, Oyo, Oyo State
Contact: [email protected] ,+2348032068784
4
Department of Computer Science, University of Ibadan, Ibadan, Nigeria.
Contact: [email protected], +2348033264855
ABSTRACT
Students’ academic performance prediction model is a model developed to reduce the students’
failure rate in higher institutions of learning. The model is developed to ascertain the students’
academic performance looking at the extent of socio economic background on the student’s
performance and the effect of student’s previous academic status (GPA) on his/her future
performance.
This paper proposed the J48 Decision tree model for predicting the academic performance of
third year Bachelor of Science (B. Sc) students in Computer Science of Ajayi Crowther
University, Oyo as they proceed to the final year. The data collected from the department contain
the students’ Name, Matriculation number, Gender, State of Origin, Current G.P.A. The
parameter used as the Key Performance Indicator (KPI) is the Current G.P.
The model is created using Training data set, the performance confirmed by the Supplied test set
and applied to the students’ data to generate the Prediction model. The extracted knowledge from
the prediction model is used to identify and profile the student to determine the students’ level of
success in the final year. The outcome of the work showed that the students have a high tendency
to have the same GPA in the final year.
Keywords: Data mining, Educational Data Mining, Key Performance Indicator, Prediction,
Decision tree, WEKA, Classification.

1. INTRODUCTION
Education is an important factor which contribute immensely to the growth of the society at
large. Many educational organizations and school administrations today, leave no stone unturned
to improve their student’s academic performance, in which the marks obtained by the student in
the examination decide his/her future. They want to increase the number of students getting
passed yearly in order to develop the best quality of the education process in their institution, to
maintain the brand name of the organization and to educate students in a better way.
To achieve this target, Data Mining (DM) is considered as one of the most suitable technique in
giving additional insights to the Institutions of Higher Learning community to help them make
better decisions in educational activities. Data mining is the process used to transform data to
useful information by extracting and making use of relatively unknown patterns, trends and
dataset from large amount of stored raw data in the repository. Though many factors determine a
good institution, the academic performance of the student play a vital role in it (Jai & David,
2014).
Educational Data Mining (EDM) is an emerging trend in data mining, which is concerned with
developing techniques for exploring and mining useful educational patterns/dataset from the
database of an educational institution and using the extracted data to make decisions and
predictions for the enhancement of the educational system as it relate to students’ performance
and systemic improvement. Researchers in this field focus on discovering useful knowledge
either to help the educational institutes manage their students better, or to help students to
manage their education and deliverables better and enhance their performance (Amjad, 2016).

2. LITERATURE REVIEW
Data Mining
According to Mehmed (2011), Data mining is a process of discovering various models,
summaries, and derived values from a given collection of data. It has been widely used in recent
years due to the availability of huge amounts of data in electronic form, and there is a need for
turning such data into useful information and knowledge for large applications. These
applications are found in fields such as Artificial Intelligence, Machine Learning, Market
Analysis, Statistics and Database Systems, Business Management and Decision Support
Systems.
In practice, the two primary goals of data mining tend to be prediction and description (Mehmed,
2011). Prediction involves using some variables or fields in the data set to predict unknown or
future values of other variables of interest. Description, on the other hand, focuses on finding
patterns describing the data that can be interpreted by humans.
Therefore, it is possible to put data - mining activities into one of two categories:
1. Predictive data mining, which produces the model of the system described by the given data
set, or
2. Descriptive data mining, which produces new, nontrivial information based on the available
data set.
On the predictive end, the goal of data mining is to produce a model, expressed as an executable
code, which can be used to perform classification, prediction, estimation, or other similar tasks.
On the descriptive end of the spectrum, the goal is to gain an understanding of the analyzed
system by uncovering patterns and relationships in large data sets.

Data mining techniques

Data mining techniques are used to build a model to identify new knowledge information
(Osmanbegovic and Mirza, 2012). Several major data mining techniques have been developed
and used including association, classification, clustering, prediction and sequential patterns.
Mehmed (2011) postulated that the goals of prediction and description are achieved by using
data-mining techniques for the following primary data-mining tasks:
1. Classification - Discovery of a predictive learning function that classifies a data item into one
of several predefined classes.
2. Regression - Discovery of a predictive learning function that maps a data item to a real - value
prediction variable.
3. Clustering - A common descriptive task in which one seeks to identify a finite set of categories
or clusters to describe the data.
4. Summarization - An additional descriptive task that involves methods for finding a compact
description for a set (or subset) of data.
5. Dependency Modelling - Finding a local model that describes significant dependencies
between variables or between the values of a feature in a data set or in a part of a data set.
6. Change and Deviation Detection - Discovering the most significant changes in the data set.

Data mining process

The general experimental procedure adapted to data – mining problems involves the following
steps:
1. State the problem and formulate the hypothesis
2. Collect the data
3. Pre-process the data
4. Estimate the model
5. Interpret the model and draw conclusions

Educational Data Mining

According to Amjad (2016), Educational Data Mining (EDM) is a new trend in the data mining
and Knowledge Discovery in Databases (KDD) field which focuses in mining useful patterns
and discovering useful knowledge from the educational information systems, such as, admissions
systems, registration systems, course management systems (module, blackboard, etc…), and any
other systems dealing with students at different levels of education, from schools, to colleges and
universities.
Researchers in this field focus on discovering useful knowledge either to help the educational
institutes manage their students better, or to help students to manage their education and
deliverables better and enhance their performance. It is an emerging discipline, concerned with
developing methods for exploring the unique types of data that come from educational settings
and using those methods to better understand students and the settings which they learn (Kumar
and Radhika, 2014).
In this paper the Traditional Class Room Educational System is considered. The performance of
the students depends on many factors such as their family background. Also important is the
academic background i.e. Secondary School Certificate Examination (SSCE), all previous
semesters’ results and Cumulative Grade Point Average (CGPA) has to be considered. Those
students who are on risk can be find out and some remedial action can be taken to improve their
performance. Due to this results the institution can improve chance(s) of the students who will
pass. The educational data mining can be used to get feedback for the teachers so the teacher can
improve the teaching method. Also the information can be useful for those who are designing the
course contents.

Key Performance Indicator

Osmanbegovic and Mirza (2012) presented a model to predict student performance. They
evaluate student success by passing grade at the exam. Parameters addressed for prediction
include students' socio-demographic variables, achieved results from high school, the entrance
exam, and attitudes towards studying which can have an effect on success.
Ramesh et al. (2013) presented a valuable study to figure out factors that influenced student
success. They focused on parents’ occupation and school type. Their obtained results from
hypothesis testing reveals that type of school does not influence student performance and parents'
occupation plays a major role in predicting grades. The ability to predict students’ mark could be
useful in a great number of different ways associated with university-level learning.
Even though there are many prediction model available with different approaches in student
performance, Sajadin (2011) posited that there is no certainty if there are any predictors that
accurately determine whether a student will be an academic genius, a drop out, or an average
performer.
Graded Point Average (GPA) is a commonly used indicator of academic performance. Many
universities set a minimum GPA that should be maintained. Therefore, GPA still remains the
most common factor used by the academic planners to evaluate progression in an academic
environment. Many factors could act as barriers to student attaining and maintaining a high GPA
that reflects their overall academic performance, during their tenure in university. These factors
could be targeted by the department in developing strategies to improve student learning and
improve their academic performance by way of monitoring the progression of their performance
(Oyelade et al., 2010).
In this work, the key Performance Indicator used is the Graded Point Average (GPA) of the
student, bearing in mind that the student’s family background and SSCE result cannot be fully
relied upon as key performance indicator to predict his/her future performance.

The Waikato Environment for Knowledge Analysis (WEKA)

According to Frank et al. (2005), WEKA is a data mining system developed by the University of
Waikato in New Zealand that implements data mining algorithms. WEKA is a state-of-the-art
facility for developing machine learning (ML) techniques and their application to real-world data
mining problems. It is a collection of machine learning algorithms for data mining tasks. The
algorithms are applied directly to a dataset. WEKA implements algorithms for data pre-
processing, classification, regression, clustering, association rules; it also includes a visualization
tools.
WEKA would not only afford a toolbox of learning algorithms, but also a framework inside
which researchers could implement new algorithms without having to be concerned with
supporting infrastructure for data manipulation and scheme evaluation. WEKA is open source
software issued under General Public License (Pallavi, 2015).
The data file normally used by WEKA is in Attribute Relation File Format (.arff), which consists
of special tags to indicate different things in the data file foremost: attribute names, attribute
types, attribute values and the data. For working of WEKA we do not need the deep knowledge
of data mining that is the reason why it is a very popular data mining tool. It also provides the
Graphical User Interface of the user and provides many facilities. The GUI Chooser consists of
four buttons—one for each of the four major WEKA applications. The buttons can be used to
start the following applications:
Explorer: It is the main interface in WEKA. It has a set of panels, each of which can be used to
perform a certain task. Once a dataset has been loaded, one of the other panels in the Explorer
can be used to perform further analysis.
Experimenter: An environment for performing experiments and conducting statistical tests
between learning schemes.
Knowledge Flow: This environment supports essentially the same functions as the Explorer but
with a drag – and drop interface. One advantage is that it supports incremental learning.
Simple CLI: Provides a simple Command-Line Interface that allows direct execution of WEKA
commands for operating systems that do not provide their own command line interface.
A trial version of WEKA package can be downloaded from the University of Waikato website
athttp://www.cs.waikato.ac.nz/~ml/weka/index.html.

Figure 1: WEKA tool front view (Source: Purva and Kamal, 2015)

Execution in WEKA
Execution in WEKA is a step by step process (Purva and Kamal, 2015). First is data loading,
Data can be loaded from various sources, including files, URLs and databases. WEKA has the
capacity to read in ".csv" format (Comma Separated Value). Firstly we take excel datasheet from
real world, the first row contains the attribute names (separated by commas) followed by each
data row with attribute values listed in the same order (also separated by commas), convert
in .csv file format. Then go to the explore button in WEKA and save this .csv file. Once data is
loaded into WEKA, the data set automatically saved into ARFF format (Purva and Kamal,
2015).
Figure 2: Execution in WEKA tool (Source: Purva and Kamal, 2015)

Decision Tree Classification in WEKA

Classification is a process of finding a model that describes and distinguishes data classes and
concepts in order to predict the class of objects whose class label is unknown. The derived model
should be represented in decision tree or neural networks.
The Classification process involves the following steps:
1. Create training dataset.
2. Identify class attribute and classes.
3. Identify useful attributes for classification (Relevance analysis).
4. Learn a model using training examples in Training set.
5. Use the model to classify the unknown data samples.
Classification is possibly the most frequently used data mining technique. It is the process of
finding a set of models that describe and differentiate data classes and concepts, for the purpose
of being able to use the model to predict the class whose label is unknown. There are many
algorithms that can be used for classification, such as decision trees, neural networks, logistic
regression, etc. and various decision tree algorithms are used in classification like ID3, AD Tree,
REP, J48, FT Tree, LAD Tree, decision stamp, LMT, random forest, random tree etc. In this
work we are using J48 decision tree algorithm for classification.

3. METHODOLOGY
In this thesis, the students’ Grade Point Average (GPA) is selected as a dependent parameter.
The dataset collected of forty-nine (49) instances (students) was used to develop a training model
whose precision level and other parameters were considered to ascertain the model’s accuracy.
The model’s status was further confirmed by the User supplied Test set which was supplied to
the system, and the statistic of the result was compared to the Training model to confirm the
accuracy of the model, this was done by supplying the model the result of the first nine (9)
students and comparing the result generated.
To generate the prediction model, a class of unknown result (Predicted GPA) was created and
supplied to the model. The prediction level of the result and other performance metric signifies
the accuracy of the prediction for each student as revealed.
Algorithm of the model
The training dataset (I1…In) which is the total number of instances was loaded to WEKA. The
GPA (Key Performance Indicator) for each student (GPA1…GPAn) from the dataset with other
information about each student are used to build the model, data of unknown GPA (nG1…nGn)
for each student was fed into the model for prediction. The result generated displays the Previous
GPA and the Predicted GPA.
Step 1 - Start
Step 2 - Take input which is given by User
In= {I1,……...,In}
Step 3 - Dataset preparation
Dn = {{I1,…In}D}
Step 4 - Dataset elaboration
DI = {GPA1,…GPAn, I1,…..In, nG1,…..nGn}
Step 5 - Processing
While (Dn!=0 )
{
If (nGn==In)
Check GPAn;
}
Step 6 - Result Generation
R = {GPA, nG};
Step 7 - Stop
Where,
In = Input given by users
Dn = Dataset
D = Database
DI = Dataset contents
GPA1….GPAn = Previous GPA for each student
nG1…..nGn = New GPA (Predicted) for each student
R = Result generated
Flowchart of the Model
The system flowchart for training the data is shown in the Figure below:
Start
 Select the file from database
Pre-process the data
Convert the file to CSV format

Pre-process selected data

CSV of pre-processed data
Classify the data using the J48 Decision Tree algorithm
Applying the classification
Process the result
Processing the result
Decision tree is produced

Stop

Figure 3: System flowchart for training of the data

Method of Data Collection

The method used for the collection of data was by getting student’s basic information such as the
Matriculation number, Name of the student, Sex of the student, State of origin and previous
academic status of the student. The data collected from the department are in a word format. The
data was copied to the Microsoft Excel (spreadsheet) format:

Figure 4: Microsoft excel format of the collected data

And saved as “comma separated value” format i.e.csv.

Figure 5: CSV format of the collected data

The data are loaded in csv format to the WEKA tool. During the scan of the data, WEKA
computes some basic statistics on each attribute. The following statistics are shown in ‘Selected
attribute’ box on the right panel of ‘Pre-process’ window:

Figure 6: The Pre-process panel showing the attributes and basic statistics

Techniques for Data Analysis

The technique used for data analysis consist of the Training phase, Testing phase, Prediction
phase and Knowledge representation. The data set are divided into two (2) sets, a training set and
a testing set. The training set is used to build the model from the J48 decision tree classifier, it
evaluates the classifier on how well it predicts the class of the instances it was trained on as the
students Current G.P.A was used as the Key Performance Indicator and various statistics in the
output such as Precision, F-measure, TP, FP are analysed. A total of forty-nine instances (49
students) were trained. The testing set is used to validate the model by checking the accuracy of
the model on how well it predicts the class of a set of instances loaded from a file as user
supplied test set. The prediction phase consist of the Prediction model created by leaving a blank
column in which the prediction level is inserted by the classifier from which the students’
performance is predicted based on the level (ranging between 0.5 to 1).. The result generated by
the model can further be represented in visualized (chart etc.) or tree form for further
clarification and easy understanding.

4. RESULTS AND DISCUSSION

This work was developed using the J48 decision tree classifier in WEKA to predict the academic
performance of forty-nine (49) students of the department of Computer Science under seven (7)
attributes.
From the experimental results, J48 algorithm predicted the unknown category of CGPA to the
accuracy of 93.8776% which is fair enough for the system to rely upon for the prediction of
students’ academic performance and it also takes relatively little time to execute.
Setting Test Options
Two categories of data set were used in the course of the work, they are:
1. Training data set - evaluates the classifier on how well it predicts the class of the instances it
was trained on. To produce the model, a training data was used, a data set with known output
values was used to build the model. This type of model used the entire dataset as the training set
which was used to create the model and use the same dataset to test the accuracy of the model
having identified the unknown attribute.
The summary of the model is as shown below:
=== Summary ===
Correctly Classified Instances 46 93.8776 %
Incorrectly Classified Instances 3 6.1224 %
Kappa statistic 0.9373
Mean absolute error 0.0027
Root mean squared error 0.0365
Relative absolute error 6.2624 %
Root relative squared error 25.0284 %
Total Number of Instances 49

The correctly classified instances are 46 at 93.8776% which signifies that the model is efficient
enough to be relied upon in contrast to the incorrectly classified instances. The Kappa statistics is
0.9373 or 93.73% which indicate that the classifier is doing better than mere chance.
2. Supplied test set - evaluates the classifier on how well it predicts the class of a set of
instances loaded from a file. Clicking on the ‘Supplied test Set’ button brings up a dialog
allowing you to choose the file to test on. The test data was created to control over-fitting after
the model is created, it is tested to ensure that the accuracy of the model does not decrease with
the test set. This ensure that the model will accurately predict future unknown values. Only nine
(9) instances (the first nine students) were selected and used to evaluate the model.
=== Summary ===
Correctly Classified Instances 9 100 %
Incorrectly Classified Instances 0 0 %
Kappa statistic 1
Mean absolute error 0
Root mean squared error 0
Relative absolute error 0 %
Root relative squared error 0 %
Total Number of Instances 9
Ignored Class Unknown Instances 40
=== Detailed Accuracy By Class ===

TP Rate FP Rate Precision Recall F-Measure MCC ROC Area PRC Area Class
1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 2.85
1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 2.96
1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 4.04
1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 4.77
1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 4.88
1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 3.92
1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 2.65
1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 3.35

A total of nine (9) instances (students) was used to evaluate/test the model. The model have the
same information as the Training data set model, thus showing that it is okay and efficient for
prediction.
The correctly classified instances are 9 at 100%, the incorrectly classified instances 0 at 0%,
Kappa Statistics is 1, Ignored instances 40 (this is because only nine (9) students were used to
test the model out of the forty nine (49) students of ACU), True positive (TP) is 1, False positive
(FP) is 0, Precision Rate, Recall and F-Measure are 1. It thus reflects that the model strong and
better for prediction.

Prediction Model
After training and testing the model, data of unknown academic performance was then fed into
the system for prediction. The prediction output of a given student is either “Pass” or “Fail”. Out
of forty-nine (49) datasets supplied into the system forty-six (46) were correctly predicted and
only three (3) were incorrectly predicted. The percentage of correctly predicted dataset is
93.8776% which is fair enough to be entirely depended on.
=== Summary ===
Correctly Classified Instances 46 93.8776 %
Incorrectly Classified Instances 3 6.1224 %
Kappa statistic 0.9373
Mean absolute error 0.0027
Root mean squared error 0.0365
Relative absolute error 6.2624 %
Root relative squared error 25.0284 %
Total Number of Instances 49

The model having correctly classified instances of 46, Kappa statistics of 0.9373, True positive
rate of 1, False positive rate of 0, Precision rate and Recall of 1 and F-measure of 0.667-1, shows
that most of the students have a high tendency to have the same GP in the coming Semester.
The table below shows and interpret the prediction of the first nine (9) students as predicted by
the model:
Student Name Matric. Current Predicted Prediction
Number G.P G.P Level
ADENIRAN IfeoluwaAdebola 15N02/001
2.85 2.85 1

ALAO Ibukunoluwa 15N02/002

2.96 2.96 1

ATUNWA MichealAdedamola 15N02/003

4.04 4.04 1

AYINLA Kwamdeen 15N02/004

4.77 4.77 1

BandeleOlaomopoOluwanifemi 15N02/005
4.88 4.88 1

ODERINDE OluwadamilareAdiyo 15N02/007

3.92 3.92 1

OGUNWOLE AdeyemiOlusola 15N02/008

2.65 2.65 1

OYATUNDE Damilare 15N02/010

2.96 2.96 1

OYENUSI Adetayo Samuel 15N02/011

3.35 3.35 1
 6
5
4
3
2
1
0
la a a n i yo la re l
bo uw ol ee f em di so i la ue
l m d i A u m
de no da am an re
l m Sa
a
A ku de w w la iO Da yo
u A K l u i m E a
uw Ib l A O m ye D t
ol AO ea NL po da de UN de
Ife L i ch Y I o a A T I A
A A m uw E A S
RA
N
A
M
l ao Ol OL OY NU
I W O E W E
EN UN le ND N OY
AD AT n de E RI O GU
Ba OD

Current G.P Predicted G.P

Figure 7: Prediction graph for the first nine students

From the graph, the Current G.P is represented by the blue bar (left) while the Predicted G.P is
represented by the orange bar (right). The graph shows that the Predicted G.P and the Current
G.P are at the same level for all the students, thus signifying that the student is liable to have the
same G.P in the coming Semester. For example Adeniran Ifeoluwa had a G.P of 2.85 is likely to
have the same G.P in the coming Semester because the prediction level is 1, the same goes for
other students.

Summary
In this thesis, J48 method of classification was used to build a model to predict the students’
academic performance on the basis of its accuracy measure and prediction level for small
datasets, and in terms of its merits above other decision tree algorithms. As the model was
evaluated using student’s record of the Computer Science Department, Ajayi Crowther
University, Oyo.
The simplicity of the J48 result output and its easy interpretation and prediction makes it a more
convenient tool to predict students’ academic performance.

Conclusion
The J48 decision tree present and achieve a high rate of accuracy. It classify the data into the
correctly and incorrectly instance as we cover seven (7) attributes under forty-nine (49) instances
and the model successfully identifies the students who are likely to fail. These students can be
considered for proper counselling so as to improve their result in the coming Semester.
The system will generally help students to benchmark their grades from their entry point into the
final year, thereby helping them to work harder in order to achieve this. Finally, the developed
system would help to significantly reduce the overall failure rate in most academic institutions as
students can be well guided and counselled.
Recommendation
From the results and findings of the experiments in this work, we recommend the adoption of
student performance prediction model as Education Data Mining is an emerging data Mining
discipline. More similar studies on different data set for machine learning approach is needed to
confirm the above finding.
The future work could also include applying data mining techniques on an expanded data set
with more unique attributes to get more accurate results. Also, a comparative analysis of these
results would be carried out based on other experiments results gotten from using other types of
decision tree algorithms such as C4.5, CHAID and CART.
REFERENCES
A. Dinesh Kumar, Dr. V. Radhika, A Survey on Predicting Student Performance,
International Journal of Computer Science and Information Technologies (IJCSIT), Vol. 5 (5),
2014, 6147-6149.

Amjad Abu Saa (2016), Educational Data Mining & Students’ Performance
Prediction, International Journal of Advanced Computer Science and Applications (IJACSA),
Vol. 7, No. 5.

E. Frank, M. Hall, G. Holmes, R. Kirkby, B. P Fahringer, I. H. Witten, and L. Trigg (2005),

WEKA
in Data Mining and Knowledge Discovery Handbook, Springer, Pp. 1305 –1314.

Jai Ruby & K. David (2014), “A study model on the impact of various indicators in the
performance of students in Higher Education“, IJRET International Journal of Research in
Engineering and Technology, Vol. 3, Issue 5, May 2014, Pp.750-755.

Mehmed Kantardzic (2011), Data Mining: Concepts, Models, Methods, and

Algorithms, Second Edition, Institute of Electrical and Electronics Engineers. Published 2011 by
John Wiley& Sons, Inc.

Osmanbegovic E, Mirza S (2012), Data mining approach for predicting student performance.
JEcon Bus 10: 3-12.

Oyelade, Oladipupo & Obagbuwa (2010),”Application of K-means clustering algorithm for

prediction of student’s academic performance.” IJCSIS, Vol.7, No.1, Pp-292.

Pallavi, Sunila Godara, (2015) A Comparative Performance Analysis of Clustering

Algorithms, International Journal of Engineering Research and Applications (IJERA) ISSN:
2248-9622www.ijera.com Vol. 1, Issue 3, Pp. 441-445.

Purva Sewaiwar, Kamal Kant Verma (2015), Comparative Study of Various Decision
Tree Classification Algorithm Using WEKA, International Journal of Emerging Research in
Management &Technology ISSN: 2278-9359, October, 2015 (Volume-4, Issue-10)

Ramesh V, Parkavi P, Ramar K (2013), Predicting student performance: A statistical and data
mining approach, IJCA 63: 35-39.
Sajadin et al (2011), Prediction of student academic performance by an application of
data mining techniques, International Conference on Management and Artificial Intelligence
IPEDR Vol.6 IACSIT Press, Bali, Indonesia.