International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)

Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

STUDENT’S PERFORMANCE PREDICTION

USING MULTIPLE LINEAR REGRESSION AND
DECISION TREE
Ojo Grace Funmilayo, Afolabi Ibukun.T.,
Computer and Information Sciences, Covenant University, Ota, Ogun State
Computer and Information Sciences, Covenant University, Ota, Ogun State

Abstract— Currently, there is rapid growth and development in the educational sector. This growth emerges
from the current technologies and the procedure that is directed towards improving student performance. In
order to contribute to this development, various factors are being used to optimize academic performance and
improve student retention. This study aims at predicting student performance at an early stage of the
academic program and monitoring students with poor academic performance. The early performance
prediction is targeted towards managerial advising based on student recruitment and the monitoring aspect is
directed for academic level advising and intervention. The methodology of this research is in three major
phases. The first phase is the data pre-processing phase which captures data visualizations using Bar charts,
histogram, box plots and scatter plots. This phase also includes algorithm such as Exhaustive Search for data
dimensionality reduction. The second phase captures the application of the machine learning algorithm used
for the student performance prediction and evaluation. Multiple Linear Regression and Decision Tree C4.5
was used and evaluated using precision and recall. In the final phase, a proof of concept was created in form
of an online application which is able to give managerial advice and academic level advising. This
application was implemented using the python programming language. The results obtained is able to assist
in recruitment and level advising with a prediction model accuracy of 60.53%. Furthermore, the application
developed is able to provide explanation which helps in diagnosing the unique challenges of individual
students.

Keywords— Academic Performance, Multiple Linear Regression, Decision Tree, Prediction, Data Mining,
Level Advising.
1. INTRODUCTION

Today, one of the biggest challenges in higher education institutions is the proliferation of data and how to
critically utilize these data to improve quality academic programs and services for decisions making [1], [2]
Data mining applications are more frequently used particularly in solving educational and administrative
problems in higher education. Therefore, Data mining is traditionally known as knowledge discovery to uncover
“hidden information,” patterns and knowledge within large volume of data hence it is the practice of making
predictions for outcomes or behaviors [3]. Data mining can best be seen as an automatic process of getting out
useful knowledge and information including patterns, associations, changes, trends, anomalies, and significant
structures that are unknown from a large chunk of complex datasets [4]. At the moment, educational institutions
gather massive amount of data that involves information regarding student enrolment, attendance records, as
well as their examination results. Extracting such nature of data brings about interesting information that serves
its users well. These large data stored in the educational database warehouses and libraries requires developing
more powerful packages for analysis and transformation of interesting knowledge by extracting information that
could help in decision making [5].

In the recent time, data mining techniques has been used extensively in higher education. This has been
necessitated by its huge potentials in educational institutes especially in better allocation of resources, academic
planning based on intervention transfer prediction, improved effectiveness of alumni development [6], student
academic performance prediction [7], emergence of new field called educational data mining (EDM) [8], [2].
The EDM process transforms raw data coming from educational systems into meaningful and useful
information that could probably have a better impact on educational enquiries and practice [9].
Practically, EDM brings about additional and detailed insights to educational entities such as lecturers, students,
alumni, staff, and managerial behavior [10]. In essence, it can ultimately be used to carry out both staff and
resources allocation in an effective way. This helps to make better decisions on various educational activities in
order to improve students’ success, increase students’ learning outcome, and other factors that determines
student success and/or failure rate. [10], [11]. Predicting students' academic achievements is one of the most
popular applications in Educational Data Mining. In the current trends, student performance plays a tremendous

All Rights Reserved © 2019 IJARCET 256

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

role in every educational sector by promoting good and excellent report of the institution and if any institution
lacks excellent performance, it will drag the good image of the institution into the mud [12], since academic
success is one of the predominant factors taken into consideration by the organization in recruiting workers
particularly the fresh graduates as a result of students performance predicted due to the effort anticipated by the
student in their study to acquire good grade to meet with the employer’s demand [13].

There are various data mining techniques processes used for drawing out meaningful knowledge from student
data which involves different classification tasks such as decision tree algorithms, Bayesian Classification, and
classification by Back Propagation, Support Vector Machine and K Nearest Neighbour. Decision tree algorithms
are used for gaining information for the purpose of decision-making and is one of the most widely used
algorithms in data mining. The decision tree starts with a root node from which users take actions, from this first
node, users split further nodes recursively according to the decision tree learning algorithm [14].

Decision Tree is by way one of the most prevalent methods of predicting student performance. This is because
of its simplicity and comprehensibility in uncovering small or large data structure and predict the value in a
clear and meaningful way [15] and the method used is the multiple linear regression analysis.

In this research, the recruitment dataset of student within the period of thirteen years from 2005/2006 to
2017/2018 academic session of Computer Science & Information department at Covenant University, Ota,
Nigeria were gathered for a total of 991 enrolled student during the period of different attributes. The work uses
Multiple Linear Regression analysis and Decision tree e.g. C4.5 to develop the model for student performance.
The techniques used for the model selection operates on two different report; one for recruitment and the other
for level advising. This research work aims to develop student academic performance predictive system using
multiple linear regression and decision tree algorithm.

II. RELATED WORK

In the educational System, Prediction of student academic performance has been applied to various schools and
colleges for improving student performances hence, managerial skills is achieved. In Educational Data mining,
various techniques are used to improve student performances in future. The use of dataset from the academic
domain, educational data mining algorithms are also introduced to predict and improve student performance in a
module of automated intelligent education systems. The major tasks for predicting student performance is by
Classification and algorithm used are Decision tree, Artificial Neural Networks, Naive Bayes, K-Nearest
Neighbor and Support Vector Machine among several other. Here are the outcomes of some of the researchers
in this field and how they were able to accomplish these tasks
El-halees [16] presented a case study that used educational data mining to analyze students’ performance. In the
study, educational data mining was applied to improve graduate students’ performance, and overcome the
problem of low grades of graduate students. It was discovered that after preprocessing the data, data mining
techniques were utilized to discover association, classification, clustering and outlier detection rules. The four
tasks analyzed the extracted knowledge and describe its importance in educational domain.
Singh and Kumar [17] applied classification techniques to predict student performance and six classification
methods were used on student data i.e. Multilayer Perceptron, BayesNet, Naïve Bayes, IB1, Decision table and
PART Classification. It was recorded in the analysis that IB1 Classifier is most suitable techniques used to
derive best performance for the categorical type of student dataset. The performance analysis of the student’s
academic records and check which algorithm is optimal for categorizing students’ based on their final grade.
This analysis was as well used to classifies the performance of Students into Excellent, Good and Average
categories.
Mayilvaganan and Kalpanadevi [18] focused on the comparison of Classification techniques for predicting
students’ academic performance and this analysis is done to checkmate the skills of the slow learners. The
techniques used are C4.5 algorithm, AODE, Naive Bayesian classifier algorithm, Multi Label K-Nearest
algorithm to find the best accuracy of classification algorithm and decision tree algorithm.
Abu-oda and. El-Halees [19] focused on the implementation of data mining techniques and methods for
acquiring new knowledge from data collected by universities to discover the dropout student hence, student
enrollment data, CGPA, major courses and Results were used to predict student performance by applying
Classification Algorithm such as Decision Tree (DT), Naive Bayes (NB), FP-growth algorithm.
Shanmugarajeshwari and Lawrance [20] analyzed the use of classification techniques for evaluating student
performance and improve students’ performance. Feature selection and classification techniques were used with
Decision tree induction and Naïve Bayes algorithm to determine student dropout.
Sagardeep and Anchal [21] Improves on student performance by applying Naïve Bayes classifier, J48 Decision
Tree and Multilayer Perceptron (MLP) to predict student’s academic performance and identifying students who
are likely to drop out, weak and declined drastically from improving academically. WEKA (Waikato
Environment for Knowledge Analysis) was used as the software tools written in JAVA for pre-processing,
classification and visualization of the data.
All Rights Reserved © 2019 IJARCET 257
 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

III. METHODOLOGY
There are four major processes to the methodology used in this paper which includes data gathering and feature
selection, data visualization, developing and evaluation of model using multiple Linear Regression and C4.5,
develop a web interface to make prediction for recruitment and provision of explanation module for proof of
concept. In this research, our first goal is to gather data from the student recruitment dataset since 2005/2006 to
2017/2018 academic session of Computer Science & Information department at Covenant University, Ota,
Nigeria and the student records comprises a total number of 991 was used for analysis. There will be two (2)
different reports to be achieved, one for recruitment and the second one is for level advising. The methodology
workflow is presented in Figure 1.
Methodology Workflow

Data Collection from Covenant University Student database

containing
undergraduate and
graduate student
Data Visualization Techniques
e.g Bar Charts, Histogram, Box
plots, Scatter plot

Data Dimension Reduction Data Transformation Techniques

and Correlation
and Exhaustive Search

Multiple Linear Regression Classification Algorithm C4.5 and

Model Selection and Evaluation
Evaluation

Proof of Concept
(Academic Level Advising)

Explanation Management

Evaluation of the System

(Proof of Concept)

Figure 1. Methodology Workflow

Collection of Data: This research work combine the attributes gathered for recruitment of students in
department of Computer Science and attributes from graduated class of Computer Science Department in
Covenant University within a period of years from the student archive. The dataset consists 991 record and 14
attributes, 468 record and 9 attributes respectively. Each of the attributes data are classified to be numerical,
categorical and strings data type.
Data Preprocessing and Visualization: After the data is collected from the database source called CSV files,
data preprocessing is introduced to find the incorrect values, missing values, duplicate rows and columns with
all same values, then eliminate the unnecessary values that does not contribute to student performance. For the
data collected from CSIS to be well analyzed, we decided to visualize the data using the following statistical
methods which includes Bar charts, linear graphs, scatter plots, box plots for the course of this study mostly for
selecting a particular attribute. In this process, we used scatter plot matrix to produce result for each variables
corrected and missing of values. The data attributes was then reduced using Dimensionality Reduction
techniques to identify the variables to be removed for the model to function effectively. This was analyzed:
firstly on the record dataset of 991 from the recruitment dataset since the large amount of data does not affect
the processing but it reduced the total number gathered. Secondly, The Cumulative GPA has large number with
continuous values and for effective and best performance, the attribute is discretized to categorical values from
graduated class dataset For instance, we classified the CGPA into First Class, Second Class Upper, Second
Class Lower and Third Class.
Multiple Linear Regression Analysis and Decision Tree: In this research, Multiple Linear Regression was
used for model selection and to describe the relationship between the data attributes. Regression analysis was
used to study the relationship between peformance measures in terms of (CGPA i.e L100, L200, L300, L400
and the sex, CUSATSCORE, JAMB SCORE) and also variable selection was done using Exhaustive Search
from the recruitment dataset. The reason for using Exhaustive Search approach is to determine the possibility of
a moderate number of predictors that is more suitable for the prediction. The subset attribute preferable for
All Rights Reserved © 2019 IJARCET 258
 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

multiple linear regressions illustrated in Table 1 and Table 2 shows the result of anaysis of varance (ANOVA)
and the exhaustive search for the recruitment dataset respectively. The linear regression model is developed to
predict the performance of the model based on recruitment data result. The dataset is divided into 70% for
training and 30% for testing data The Linear Regression Model is presented in Table 3. Hence, the formular for
Multiple Linear Regression is illustrated thus: ------ (3-1)
Where are coefficient and is the noise or unexplained part.
yi​ = Dependent variable
xi​ = Explanatory variables
​ = y- intercept (constant term)
βp = Slope coefficients for each explanatory variable

Prediction Accuracy Measures

The prediction error for recruitment dataset record is defined as the difference between its actual outcome
value and its predicted outcome value: . A few popular numerical measures of predictive
accuracy are:
MAE (mean absolute error/deviation) = . This gives the magnitude of the average absolute error.

Mean Error = . It retained the sign of the errors measured and negates the errors out of the positive
errors of the same magnitude.
MPE (mean percentage error) = . This gives the percentage score of how predictions
deviate from the actual values (on average) illustrates the direction of the error.
MAPE (mean absolute percentage error) = . This is used to calculate how
predictions deviates (on average) from the actual values and gives a percentage score of the prediction model. It
used for evaluating the residual error. The lower the value of the MAPE, the lower the prediction error of the
predicted model.

RMSE (root mean squared error) = . This is computed on the validation data. It has
the same units as the outcome variables
Decision Tree for searching the path in Recruitment data result: This was done to categorise the graduated
class data into First Class, Second Class Upper, Second Class Lower, Third Class and Failed. Recommendation
combination to get GOOD result. [2:1 – First Class] based on the parameter optimization to find which
combination would give a better result for the student. That is, using the predictors such as CUSATSCORE, Sex:
0 {where Male =1 and Female = 0} Age, JambScore to trace the path for decision optimization. We keep
looping the variable inputs till we get output that is satisfied not necessary optimal, so from Second class upper
to First Class and the optimization on the variable provides the prediction output of in a categorical values.
A. METRICS FOR ACCURACY
This is done by taken the percentage of the accuracy of the model to be able to ascertain the best performance
for our predictions. The following measures are used to check for the accuracy:
Precision: the fraction of true positive of the number of relevant records that was recovered to the total number
of irrelevant and relevant records recovered.
Recall is the proportion of the number of relevant records recovered to the total number of relevant records
currently in the database, examples of this is classified as positive by a classifier. Recall is the fraction of true
positive examples classified correctly by a classifier.
F-measure is the harmonic mean of the precision and recall of a classifier; that is, F-measure=2× precision×
recall/(precision+recall) [22].
The precision and recall are often calculated in percentage as illustrated below:
= ------------------- (3-2)

, = ------------------ (3-3)

= ---------------- (3-4)

All Rights Reserved © 2019 IJARCET 259

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

= Harmonic mean of recall and precision --------- (3-5)

IV. SYSTEM DESIGN AND IMPLEMENTATION

The implementation tools and development environments used are Jupyter notebook, visual studio, helper apis
and Scikit-Learn. The python programming language was used to develop the application while Html and CSS
was used for the frontend in the development of the system. SQLite was chosen as the database management
system to manage the student information because of its capabilities to store a very large data and easy to
retrieve or update data.
B. THE APPLICATION INTERFACES

Figure 2. Homepage for the application

Figure 3. Portal Login for Level Adviser

Figure 4. Prediction model for graduating class

All Rights Reserved © 2019 IJARCET 260

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

Figure 5. Prediction classification performance

C. ACTIVITY DIAGRAM
This is the workflow of activity within the system; it shows the computational processes of the activity from one
series to another. Figure 2 shows the activity diagram of the system.

Figure 6. Activity Diagram of Prediction Interface System

V. DISCUSSION AND RESULTS

To predict the performance (graduation cgpa) of students (for recommending which students to admit and also
for academic level Advising) using data containing biodata, pre admission requirements/ results and
performance in the university. Sample of the student dataset for recruitment is illustrated in Table 1.

CGPA Sex_Male Age_entry CUSat Jamb L100 L200 L300 L400

Score Score
3.292683 1 18 39 138 138 3.050847 3.634146 3.85
3.32716 1 21 38 197 3.043478 3.193548 3.289474 3.707317
2.511111 1 18 37 162 2.136364 1.638889 3.209302 3.465116
2.625698 1 18 29 144 2.481481 1.932203 2.759259 3.589744
2.760234 1 18 30 190 2.25 2.75 2.652174 3.341463
2.843023 1 18 30 230 1.878788 2.96875 3.194444 3.128205
4.770186 0 17 42 191 4.772727 4.677966 4.756098 4.923077

All Rights Reserved © 2019 IJARCET 261

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

3.420118 0 21 34 197 3.458333 3.084746 3.425532 3.897436

3.981366 1 22 40 199 4.181818 3.288136 4.463415 4.410256
3.587879 0 18 42 164 2.916667 2.830508 4.232558 4.435897
3.046784 0 17 39 215 3.772727 2.423729 3.039216 3.589744
3.494048 1 18 32 184 3.4 2.762712 3.727273 4.375
4.055215 0 20 31 177 3.791667 3.677966 4.268293 4.564103
2.27933 1 17 18 155 2.162162 2.571429 1.941176 2.469388
2.137255 1 20 26 173 1.612903 1.634146 2.46 3.093023
4.322981 1 20 43 203 4.181818 4.20339 4.317073 4.589744
Table 1: Sample of student dataset for recruitment
Multiple Linear Regression is chosen for this prediction assignment due to numeric nature of the data and its
ability to give high prediction accuracy with less than 1000 records [23].
All the numerical attributes were each visualized using the histogram and the box plot. The histogram was used
to investigate the skewness of each numeric attribute and the box plot was used to check for possible outliers in
the numeric attribute.

Figure 7.1a: Figure 7.1b: CGPA versus ageentry

Figure 7.1c: Figure 7.1d: (cgpa versus ageexit}

It was discovered that age of entry had an outlier, which was taken care of by replacing with the average age of
exit. Also due to the fact that we have the same trend with age of entry and age of exit, and their correlation is
very high, we selected only age of entry in order to avoid multicollinearity. Multicollinearity is the presence of
two or more predictors sharing the same linear relationship with the outcome variable.

Figure 7.1e freqency of CUSATSCORE Figure 7.1f: cusatscore versus KL using Boxplot
using Histogram

All Rights Reserved © 2019 IJARCET 262

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

Figure 7.1g Frequency of JAMBSCORE Using Histogram Figure.7.1h: jambscore versus KL

Figure 7.1g &Figure 7.1h. For cusatscore and Jamb score, the histogram presented a distribution which was
averagely good and not skewed towards a particular direction of the distribution of the data.

Figure 7.1i: KL showing on Histogram and Boxplots

Figure 4.5l: is the boxplots for the numeric attributes of the gpa for 100level, 200level and 300 level. All the
four attributes presented above (L100, L200, L300, L400) are most separated in the prediction category
(firstclass, secondclass, etc) indicating potentially useful predictors.
To have a clearer picture of the data, each of the numerical attributes in the dataset was visualized using the
histogram and the box plot. The purpose of this is to see if there are missing values and examine the skewness of
the data and make conclusion if there will be need to remove some attributes or perform data transformation
based on the result of their skewness and missing values. Since the attribute to be predicted is the performance
which is either the graduating cgpa or class of degree, these were the attributes that each other attributes was
compared with (on the y axis)

All Rights Reserved © 2019 IJARCET 263

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

Figure 7.1j: showing all the attributes using scatter plot matrix
In the previous Figure 7.1a above, shows the result of the bar charts of the attributes being averagely ok, but the
result of the scatter plot in figure 4.5b shows that all the attribute gave a positive correlation result with the
predicted value which is the cgpa apart from the age of entry which gave a negative correlation.

Table 2: ANOVA for regression analysis of student performance attributes.

Table 2 shows the result of ANOVA analysis based on the data attributes which explains 99% of variance of
student performance measure (R Square=0.997735. The Significant F is 0 less than 0.5 indicating a strong
regression. The regression coefficients indicate that there is a positive relationship between the student
performance (final cgpa) and the CUSATSCORE, AGE ENTERY and sex as revealed in Table 2. The p values
also show that the predictive values of the CUSATSCORE, AGE ENTERY and sex are very reliable.

All Rights Reserved © 2019 IJARCET 264

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

Table 3: Variable Selection for Linear Regression using Exhaustive Search

Table 3 shows that values of rsq and radr2 indicates all the subsets of variables presented are qualified for the
prediction having values greater than 0.8

Furthermore, predicting student performance using multiple linear regression. The linear regression model is
developed to predict the performance of the model based on recruitment data result. The dataset is divided into
70% for training and 30% for testing data. This is presented in Table 4. Linear Regression Model for prediction
performance.

Table 4: Linear Regression Model

The regression analysis result shows that the attributes used in the data attributes explain 99% of student
performance measure (R Square=0.9974). The Significant F is 0 which is less than 0.5 indicating a strong
regression. The regression coefficients indicate that the prediction model is efficient and optimum between the
predictors i.e final CGPA performance and CUSATSCORE, AGE ENTERY and sex as revealed in Table 4. The
p values also show that the predictive values of the CUSATSCORE, AGE ENTERY and sex are very reliable.

The Residual Error is plotted on the histogram to show the variation of the error in Figure 8.

All Rights Reserved © 2019 IJARCET 265

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

Figure 8: Histogram for the regression error

Decision Tree Path illustrates the optimization of the prediction performance of the student in Figure 9.

Figure 9: Decision Tree Path For Student Prediction Performance

Sample node path derived using python from the Decision Tree Paths from figure 8: Given input features for
100 level as:
Sex: 0 {where Male =1 and Female = 0}
Age: 17
Cusat Score: 18 (out of 100)
Jamb Score: 155 (out of 400)
Array transformed as: [1, 17, 18, 155]
Prediction output: Third Class.
Path to prediction:
['Rules used to predict sample 0: ',
'decision id node0 ( CUSATSCORE (= 18) <= 69.73333740234375)',
'decision id node1 ( CUSATSCORE (= 18) <= 37.5)',
'decision id node2 ( JAMBSCORE (= 155) <= 169.5)',
'decision id node3 ( CUSATSCORE (= 18) <= 30.0)',
'decision id node4 ( JAMBSCORE (= 155) <= 159.5)']

D. DESCRIPTIVE ANALYSIS FOR LEVEL ADVISING AND EVALUATION ANALYSIS

The goal of this prediction is to predict the performance (graduation class) of students (for level advising using
data containing result of several courses taken (time taken and course lecturers). The data used for this purpose
comprises of computer Science and MIS students that graduated from covenant university from 2013/2014 to
2017/2018 (4 academic sessions) which has a total record of 468 students. Few sample dataset is illustrated in
Table 5.

All Rights Reserved © 2019 IJARCET 266

 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

SESSION TTG TMC TMC EDS DLD CSC CSC LCSC CSC LCSC CIS LCIS
422 421 421 221 429 424 424 423 423 421 421
2017/2018 B A A A A A A Ayo A Oyelade A Okesola

2017/2018 B A A A A A A Ayo A Oyelade A Okesola

2017/2018 A A A A A A A Ayo A Oyelade A Okesola

2017/2018 A A A A A A A Ayo A Oyelade A Okesola

2017/2018 A A A A A A A Ayo A Oyelade A Okesola

2017/2018 A A A A A A A Ayo A Oyelade A Okesola

2017/2018 A A A B A A B Ayo A Oyelade B Okesola

2017/2018 A A A A A A B Ayo B Oyelade B Okesola

Table 5: Sample of dataset for graduated student

To calculate Precision, Recall and F1-score

The performance of the system is based on precision, recall and f1-score shown in Table 6.
= ------------------- (3-2)

, = ------------------ (3-3)

= ---------------- (3-4)

= Harmonic mean of recall and precision --------- (3-5)

Precision Recall F1-score Support

First class 0.83 0.56 0.67 9
Second class upper 0.57 0.61 0.59 28
Scond class lower 0.61 0.68 0.64 28
Third class 0.56 0.45 0.50 11
Average/total 0.61 0.61 0.60 76
Table 6: Evalaution for the performance of the system

VI. Conclusions
The developed system will improve the level of academic advising and provides a platform for academic
intervention to further progress the student academic performance when recruiting new student. The research
work present student prediction system used to predict the performance of the student during recruitment and
graduating class. Visualization tools were used to identify the uniqueness of the predictors and the selection of
data attributes having being correlated to increase the efficiency of the model. The Exhaustive search used
clearly avoid the skewness and multicollinearity of the predictors in order to achieve perfect correlation between
them. Multiple linear Regression used provides the model with better accuracy to make the prediction more
effective. Classification algorithm is applied using C4.5 to classify the predicted value for the model and the
decision tree gives more explanation weather the student is going to fail or pass at the end of the academic
performance grade. In other words, the research is done to be able to understand the model to be trained, its
accuracy of the performance, the explanation module for the interface and the usability of the proof of concept.

REFERENCES
[1] J. Ranjan and K. Malik, “Effective educational process : a data-mining approach,” J. Inf. Knowl. Manag.
Syst. Vol. 37 No. 4, Inf. Knowl. Manag. Syst. Emerald Gr. Publ., vol. 37, no. 4, pp. 502–515, 2007.
[2] C. Romero and S. Ventura, “Educational Data Mining : A Review of the State of the Art,” IEEE Trans.
Syst. MAN, Cybern. C Appl. Rev., vol. 40, no. 6, pp. 601–618, 2010.
All Rights Reserved © 2019 IJARCET 267
 International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 8, Issue 7, July 2019, ISSN: 2278 – 1323

[3] V. P. Bresfelean, M. Bresfelean, and N. Ghisoiu, “Determining Students ’ Academic Failure Profile
Founded on Data Mining Methods,” Proc. ITI 2008 30th Int. Conf. Inf. Technol. Interfaces, June 23-26,
2008, Cavtat, Croat., pp. 317–322, 2008.
[4] J. Meskens, J. Vermeulen, K. Luyten, and K. Coninx, “Gummy for Multi-Platform User Interface Designs :
Shape me , Multiply me , Fix me , Use me,” AVT08, ACM 1-978-60558-141-5, pp. 28–30, 2008.
[5] G. S. Gowri, R. Thulasiram, and M. A. Baburao, “Educational Data Mining Application for Estimating
Students Performance in Weka Environment”. IOP Conf. Ser. Mater. Sci. Eng. 263 032002
doi10.1088/1757-899X/263/3/032002. 14th ICSET-2017,
[6] J. Luan, “Data Mining and Knowledge Management in Higher Education,”. Pap Present. Annu. Forum
Assoc. Institutional Res. (42nd, Toronto, Ontario, Canada, June 2-5, 2002. 19p.
[7] B. Minaei-Bidgoli, Deborah A. Kashy, G. Kortemeyer, and W. F. Punch, “Predicting Student
Performance: An Application Of Data Mi"G Methods With An Educational Web-Based System,” 2003.
[8] E. Chandra and K. Nandhini, “Knowledge Mining from Student Data,” Eur. J. Sci. Res. ISSN 1450-216X
Vol.47 No.1 (2010), pp.156-163 © EuroJournals Publ. Inc. 2010 http//[Link]/[Link]
Knowl., vol. 47, no. 1, pp. 156–163, 2010.
[9] A. M. Abaidullah, N. Ahmed, and E. Ali, “Identifying Hidden Patterns in Students ‟ Feedback through
Cluster Analysis,” vol. 7, no. 1, pp. 16–20, 2015.
[11] M. Goyal and R. Vohra, “Applications of Data Mining in Higher Education,” CSI Int. J. Comput. Sci.
Issues, Vol. 9, Issue 2, No 1, March 2012 ISSN 1694-0814 [Link], vol. 9, no. 2, pp. 113–120,
2012.
[12] B. K. Baradwaj and S. Pal, “Mining Educational Data to Analyze Students ‟ Performance,” Int. J. Adv.
Comput. Sci. Appl. Vol. 2, No. 6, 2011, October 2011.
[13] S. K. Yadav, B. Bharadwaj, and S. Pal, “Data Mining Applications : A comparative Study for Predicting
Student’s performance,” Int. J. Innov. Technol. Creat. Eng. VOL.1 NO. DECEMBER, vol. 12, pp. 13–19,
2011.
[14] K. Bunkar, “Data Mining : Prediction for Performance Improvement of Graduate Students using
Classification,” 978-1-4673-1989-8/12 ©2012 IEEE, pp. 3–7, 2012.
[15] N. Srecko and M. Zwilling, “Expert Systems with Applications Student data mining solution –
knowledge management system related to higher education institutions,” no. May, pp. 1–8, 2014.
[16] M. M. A. Tair and A. M. El-halees, “Mining Educational Data to Improve Students’ Performance : A
Case Study Mining Educational Data to Improve Students’ Performance : A Case Study,” no. May 2014,
2012.
[17] S. Singh and V. Kumar, “Performance Analysis of Engineering Students for Recruitment Using
Classification Data Mining Techniques,” vol. 3, no. 2, pp. 31–37, 2013.
[18] M. D. Mayilvaganan, and Kalpanadevi, “Comparison of classification techniques for predicting the
performance of students’ academic environment. International Conference on Communication and
Network Technologies (pp. 113-118). IEEE. 2014.
[19] G. S. Abu-oda and A. M. El-halees, “Data Mining in Higher Education : University Student Dropout
Case Study,” vol. 5, no. 1, pp. 15–27, 2015.
[20] V. Shanmugarajeshwari and R. Lawrance, “Analysis of Students’ Performance Evaluation using
Classification Techniques,” 978-1-4673-8437-7/16/$31.00 ©2016 IEEE, 2016.
[21] R. Sagardeep and G. Anchal, “Predicting Academic Performance of Student Using Classification
Techniques,” 4th IEEE Uttar Pradesh Sect. Int. Conf. Electr. Comput. Electron. GLA Univ. Mathura, pp.
568–572, Oct 26-28, 2017.
[22] H. Sumyea et al., “Knowledge-Based Systems Predicting academic performance by considering student
heterogeneity,” Knowledge-Based Syst., no. December 2017, pp. 0–1, 2017.
[23] P. N. Sharma, M. Sarstedt, G. Shmueli, K. H. Kim, and K. Oliver, “PLS-Based Model Selection : The
Role of Alternative Explanations in Information PLS-Based Model Selection : The Role of Alternative
Explanations in Information Systems Research,” J. Assoc. Inf. Syst. 20(4), 346-397 doi
10.17005/[Link].00538, no. April 2019, 2018.

All Rights Reserved © 2019 IJARCET 268