2018 4th International Conference on Computer and Information Sciences (ICCOINS)

Student Academic Performance Prediction

by using Decision Tree Algorithm
Raza Hasan1, Sellappan Palaniappan2, Abdul Rafiez Abdul Raziff3,
Salman Mahmood4 and Kamal Uddin Sarker5
1,2,3,4 Department of Information Technology, School of Science and Engineering,
Malaysia University of Science and Technology, Selangor, Malaysia.
5 School of Informatics and Applied Mathematics,
University Malaysia Terengganu, Terengganu, Malaysia.
[email protected], [email protected], [email protected],
[email protected], [email protected]

Abstract—This work explores student’s academic systems are potentially accessible for data mining, analysis
performance using decision tree algorithm having parameters (and interpretation) and has captured the attention of higher
like Student’s Academic Information and Students Activity. We education administrators, academics, researchers and
collected records of 22 students from Spring 2017 semester, government agencies. E-Commerce technologies module has
studying in undergraduate level from Oman’s private Higher been selected for the study because it is the core module
Education Institution. Proposed work utilizes Electronic offered in all computing specializations, where the majority of
Commerce Technologies module since it is a core module offered students are enrolled. This module is usually taken by 3rd year
in every computing specialization. Furthermore, WEKA data students having a prerequisite of server-side programming.
mining tool is used to evaluate the decision tree algorithm for
Learning outcome of this module is twofold namely
discovery of student’s performance along with Moodle access
entrepreneurship venture or part of a small-medium industry
time. Simulation results demonstrate that Random Forest Tree
algorithm showed better accuracy than comparative decision tree
(SME). The depth of the data is not only determined by the
algorithms. Hence, shows good agreement for the training set amount of raw data on a learner but also by the availability of
provided. Therefore, the proposed work aid in improving contextual information [5]. This inferred information can be
student’s grades in the module. Helping stakeholders to analyze evaluated by using Educational Data Mining (EDM) which is
and evaluate the module delivery and results. Early detection and the widely used techniques in analyzing student’s performance
solution can be made both at the institutional level and module [6].
level.
B. Educational Data Mining
Keywords—component; VLE; WEKA; Data mining; EDM is to discover new patterns and useful knowledge
Classification; Decision tree; from the HEI systems. It is also known as Knowledge
Discovery in Databases (KDD). The data in nature is
I. BACKGROUND STUDY AND RELATED WORK hierarchical and the methods that encourage KDD are given
The digital footprint students leave behind with the use of below:
Virtual Learning Environments (VLE) creating large volumes
x Prediction referred to forecast and understanding of
of data. Universities and higher education institutions have a
student’s behavior. In result impacting educational
need to mine meaningful information out from that data. Data
outcomes infer from a single aspect of the data to a
mining makes it possible for institutions to apply the combination of other related data. Common techniques
techniques and discover the hidden knowledge within that
used in prediction are Classification, Regression and
information. By doing this it will improve the educational Density Estimation.
processes within the institution and enhancing students’
performance. x Clustering is a task to find data set those are grouped
together which can be split into full datasets into further
A. Understanding of Big Data
categories.
Higher Education Institution’s (HEIs) have entered the era
of ‘big data’ and are collecting large volumes of data relating to x Relationship determines to find out a large number of
their learners and the educational process. Institutions of higher variables within the datasets and discovering the
education are operating in an increasingly complex and relationship among them. The techniques used in a
competitive environment [1]. Big Data has gained much relationship are Association rule mining, Correlation
attention from the academia and the IT industry due to rise in mining, Sequential pattern mining and casual data
handheld devices where information is generated and collected mining.
at a rate that rapidly exceeds the boundary range [2]. These x Distillation recognizes, classify features of the data
vast amounts of data are stored in the student information depicting in a form of visualization for human
systems (SIS); including learner interactions with various inferences.
educational technologies such as learning/course management
systems (LMS/CMS); and in various databases such as x Model Discovery is a technique for validation the
admissions files, library records and other systems [3] [4]. The model which can be used as a component in another
extraction of the data derived from these aforementioned

978-1-5386-4744-8/18/$31.00 ©2018 IEEE ~ 1 ~

 2018 4th International Conference on Computer and Information Sciences (ICCOINS)

analysis. The methods for model discovery can be trends and pattern of students learning experience along with
prediction or clustering or relationship mining [7]. VLE [23] [24].
C. Related Work II. DATA AND PROPOSED MODEL
EDM is not a new field; several works have been done in
A. Data and Attribute Selection
the past in order to get the insights of teaching and learning
management. Recent research shows that DM techniques are The data used for this study comprises of student’s activity
implemented in the educational sector to find student’s and performance data in a level 3 module in computing degree
performance patterns from the data generated by the students. course at Middle East College. This study employs data from
These patterns can able to find out the problems that arise for Spring 2017 using a sample of 22 students who had registered
learning management [8] [9] [10]. in the said module from undergraduate level. The dataset
comprises of variables related to two categories which are as
Romero et al. [11] [12] used classification and association follows:
techniques with E-Learning system at Cordoba University. The
study suggested different attributes for predicting final marks x Student’s Academic Information (Cumulative
and association of Moodle online participation in discussion Grade Point Average (CGPA), High Risk
forums. They also used clustering techniques. Amelia et al. (student having high failure rate in the same
[13] used prediction on a web-based course and predicted the module), Term Exceed at Risk, At Risk (student
likelihood of passing a course based on the student’s work failed 2 or more modules previously), Student
carried out in the system. The study suggested genetic Success Center (SSC), Coursework 1 (CW1),
algorithm. Giannakos et al. [14] [15] used visualization Coursework 2 (CW2), End Semester Examination
technique and video-assisted course to learn about the attitudes (ESE) and Plagiarism Count).
and learning outcome for the students. The study suggested
x Students Activity (On Campus and Outside
single group time series to identify the relationship. From the
Campus access). This is the time spent by the
literature review, it is clear that prediction technique is mostly
student on Moodle in minutes from inside or
used on predicting student's performance. Classification is
outside the campus. The values were taken from
mostly used to relate the different parameters in student’s
MEC Moodle for E-Commerce Technologies in
activity either online or offline mode of study.
Fall 2017.
There are three main components in prediction which are
B. Proposed Model
parameter, method and tool on which the Student’s Academic
Performance (SAP) [9] is assessed. Sembiring et al. [16] took This section discusses the model carried out to perform this
parameters like Interest, Study behavior, Engage time, Believe, research. The classification model is used to predict student’s
Family support, Demographic and CGPA. The method used performance by using the academic information gathered from
was Smooth Support Vector Machine (SSVM), K-Means and Student Information System (SIS) and the numbers of minutes
Decision Tree. They used Rapidminer tool to predict the SAP spend on the Moodle (VLE) which can be seen in Fig. 1.
and presented 93.7% accuracy. Shana and Venkatachalam [17] Classification is a widely utilized technique for data mining. It
took parameters like Family background and Schooling is used to predict data known object’s class or category based
information. The method used was Naïve Bayes and Decision on previously learned classes from a training dataset. For
Tree. They used WEKA tool to predict the SAP and presented classification there are several algorithms are used such as
82.4% accuracy. Arsad et al. [18] took parameters like the Decision Trees, Neural Networks, Naïve Bayes etc.
Educational background, CGPA and Gender. The method they
used was Artificial Neural Network (ANN). Akinola et al. [19] SIS
took parameters like result subject obtained during
matriculation. The method used was ANN. They used Neuro Students
Shell Classifier tool to predict the SAP. Tribhuvan et al. [20] Academic
took parameters like Attendance, Aptitude, Assignment, Test, Information Prediction of
GPA and Grade. The method used was Naïve Bayes. They Classification Students
used WEKA tool to predict the SAP and presented 92% performance
Students
accuracy. Hashmia Hamsa et al. [21] took parameters like Activity
Internal marks, Sessional marks and Admission score. The
method used was Decision Tree and Fuzzy Genetic Algorithm.
Raheel Asif et al. [22] took parameters like Higher Secondary Moodle
Certificate (HSC) Examination Marks, Module Marks and
Yearly Marks. The method used was Decision Tree. They used
Rapidminer tool to predict the SAP and presented 83.65% Fig. 1. Proposed Datamining Model
accuracy. From the literature review, it is evident that for
predicting students perform two categories have to be C. Preparation
considered namely Student's Academic Information and Data mining algorithms are used to analyze and discover
Student's activity.. Both Student’s Academic Information and useful information from data, algorithms implemented in open-
Student’s activity have a good relationship in understanding the source data mining software called WEKA [25]. The ordinal
data is nominalized using the scale (Fail, Good, Average, Pass

~ 2 ~
 2018 4th International Conference on Computer and Information Sciences (ICCOINS)

and excellent). Table I illustrates the actual dataset and Table II III. RESULT AND ANALYSIS
illustrates the converted data into nominal dataset used in the This section presents the result obtained following the
experiment. methodology mentioned in Fig.1. Table III presents the
TABLE I. ACTUAL D ATASET accuracy and kappa results of 08 classifiers that have achieved
accuracy beyond the baseline. This result shows that the data
can be predicted for ESE performance with reasonable
accuracy.

TABLE III. CLASSIFIER PREDICTION ACCURACY AND KAPPA

Classifier Accuracy Kappa

Decision Tree J48 63.6364% 0.4172
REP Tree 45.4545% 0
Random Forest 100% 1
Logistic Model Tree 45.4545% 0
Hoeffding Tree 45.4545% 0
Decision Stump 50% 0.1103
Naïve Bayes 90.9091% 0.8616
SMO 100% 1
From the above information we can interpret that Random
Forest, Binary SMO is having the kappa value 1 implies
perfect agreement. Naïve Bayes having kappa value 0.86 can
also be considered as a very good agreement. Also, it was
noted that both the classifier having the kappa value 1 due to
nominal values have shown significant variance. Random
Forest Mean Absolute Error was 0.0324 as compared to Binary
SMO 0.0826. Similarly, Random Forest Relative Absolute
Error was 37.2857 % as compared to Binary SMO 95.2381 %.
Further investigating SMO it was found that Ignored Class
Unknown Instances was exceptionally high 22, making the
confusion matrix having all values 0. Table VI illustrates the
TABLE II. NOMINAL D ATASET
actual results and the results using the training set from the
model of each algorithm.

TABLE IV. PREDICTED RESULTS

Student Actual Random Forest Naieve Bayes SMO

1 Pass Pass Pass Pass
2 Pass Pass Pass Pass
3 Good Good Average Good
4 Pass Pass Pass Pass
5 Average Average Average Average
6 Average Average Average Average
7 Average Average Average Average
8 Average Average Average Average
9 Pass Pass Pass Pass
10 Pass Pass Pass Pass
11 Pass Pass Pass Pass
12 Fail Fail Fail Fail
13 Pass Pass Pass Pass
14 Average Average Average Average
15 Pass Pass Pass Pass
16 Pass Pass Pass Pass

~ 3 ~
 2018 4th International Conference on Computer and Information Sciences (ICCOINS)

Student Actual Random Forest Naieve Bayes SMO [8] Zacharoula Papamitsiou and Anastasios A.
17 Fail Fail Fail Fail
Economides, "Learning Analytics and Educational Data
Mining in Practice: A Systematic Literature Review of
18 Fail Fail Pass Fail Empirical Evidence," vol. 17, no. 4, pp. 49–64, 2014.
19 Fail Fail Fail Fail [9] Azwa Abdul Aziz, Nur Hafieza Ismail, and Fadhilah
20 Good Good Good Good Ahmad, "Mining Students' Academic Performance," Journal
of Theoretical and Applied Information Technology, vol. 53,
21 Average Average Average Average
no. 3, pp. 485-495, July 2013.
IV. CONCLUSION AND FUTURE WORKS [10] Vikas Rao Naidu, Baldev Singh, Raza Hasan, and
Ghaniya Al Hadrami, "Learning analytics for smart classroom
The study investigated 8 different classifiers algorithms that in higher education," International E-Journal of Advances in
might help the stakeholders to improve and improvise and Education, vol. 3, no. 8, pp. 356 - 362, September 2017.
early intervention to improve the results of the module and [11] Cristóbal Romero and S. Ventura, "Educational Data
enhancing student’s experience. It was found that Random
Mining: A Review of the State of the Art," IEEE Transactions
Forest, Naïve Bayes and SMO give good agreement for the
on Systems, Man, and Cybernetics, Part C (Applications and
training set. As, the kappa value of Random Forest and SMO
found very good as it was 1. For the supplied dataset and Reviews), vol. 40, no. 6, pp. 601-618, November 2010.
training used it was found Random Forest was best suitable for [12] Cristóbal Romero, Manuel-Ignacio López, Jose-
the module based on the lower mean absolute error and relative María Luna, and Sebastián Ventura, "Predicting students’ final
absolute error. This model can be used with the other modules performance from participation in on-line discussion forum,"
and can be tested for better analysis and accuracy in order to be Computers & Education, vol. 68, pp. 458-472, October 2013.
applicable. This can help faculty members handling a module [13] Zafra Amelia, Romero Cristóbal, and Ventura
to check the possible outcome of students in the module and do Sebastián, "DRAL: a tool for discovering relevant e-activities
the necessary actions. Stakeholders can benefit and analyze and for learners," Knowledge and Information Systems, vol. 36,
evaluate the module delivery and results. no. 1, pp. 211–250, July 2013.
[14] M. Giannakos, K. Chorianopoulos, and N.
ACKNOWLEDGMENT Chrisochoides, "Making sense of video analytics: Lessons
The authors would like to thank Middle East College, learned from clickstream interactions, attitudes, and learning
Oman for the experimental data. The authors are thankful to outcome in a video-assisted course," The International Review
Head of Computing Department Dr. Mounir Dhibi (MEC), for of Research in Open and Distributed Learning, vol. 16, no. 1,
his support and encouragement to carry out this study. 2015.
[15] Michail N. Giannakos, John Krogstie, and Trond
Aalberg, "Toward a Learning Ecosystem to Support Flipped
REFERENCES Classroom: A Conceptual Framework and Early Results In: Li
[1] Daniel Ben, "Big Data and analytics in higher education: Y. et al. (eds) State-of-the-Art and Future Directions of Smart
Opportunities and challenges," British Journal of Educational Learning," in Lecture Notes in Educational Technology.
Technology, vol. 46, no. 5, pp. 904–920, 2015. Singapore: Springer, 2015.
[2] Nawsher Khan et al., "Big Data: Survey, [16] Sajadin Sembiring, M. Zarlis, Dedy Hartama,
Technologies, Opportunities, and Challenges," The Scientific Ramliana S, and Elvi Wani, "Prediction of Student Academic
World Journal, p. 18 pages, 2014. Performance by an Application of Data Mining Techniques,"
[3] A. Atif, Richards. D., A. Bilgsin, and M. Marrone, in International Conference on Management and Artificial
"A Panorama of Learning Analytics Featuring the Intelligence IPEDR, vol. 6, Bali, Indonesia, 2011.
Technologies for the Learning and Teaching Domain.," in [17] J. Shana and T. Venkatachalam, "Identifying Key
Electric Dreams - 2013 ascilite conference, Sydney, 2013, pp. Performance Indicators and Predicting the Result from Student
68-72. Data," International Journal of Computer Applications, vol.
[4] M. A. Tair and A. El-Halees, "Mining Educational 25, no. 9, pp. 45-48, July 2011.
Data to Improve Students’ Performance: A Case Study. ," [18] P. M. Arsad, N. Buniyamin, J. L. Ab Manan, and N.
International Journal of Information and Communication Hamzah, "Proposed academic students' performance
Technology Research, vol. 2, no. 2, pp. 140-146, 2012. prediction model: A Malaysian case study," in 3rd
[5] MH Qasem, R Qaddoura, and B Hammo, International Congress on Engineering Education (ICEED),
"Educational Data Mining (EDM): A Review," in New Trends Kuala Lumpur, 2011, pp. 90-94.
in Information Technology, 2017, p. 149. [19] O. S. Akinola, B. O. Akinkunmi, and T. S. Alo, "A
[6] Amirah Mohamed Shahir, Wahidah Husain, and Data Mining Model for Predicting Computer Programming
Nur’aini Abdul Rashid, "A Review on Predicting Student's Proficiency of Computer Science Undergraduate Students,"
Performance Using Data Mining Techniques," Procedia African Journal of Computing and ICT, vol. 5, no. 1, pp. 43-
Computer Science, vol. 72, pp. 414-422, 2015. 52, 2012.
[7] Baker R, "Data Mining for Education," International [20] Tribhuvan A.P., Tribhuvan P.P., and Gade J.G.,
Encyclopedia of Education, vol. 7, pp. 112-118, 2010. "Applying Naive Bayesian Classifier for Predicting

~ 4 ~
 2018 4th International Conference on Computer and Information Sciences (ICCOINS)

Performance of a Studen," Advances in Computational using Blended Learning," in 2015 Science and Information
Research, vol. 7, no. 1, p. 239, 2015. Conference (SAI), London, 2015, pp. 797-800.
[21] Hashmia Hamsa, Simi Indiradevi, and [24] Syed Imran Ali Kazmi, Muhammad Sohail Hayat,
J.Kizhakkethottam Jubilant, "Student Academic Performance Raza Hasan, and Vishal Dattana, "Network simulation tool
Prediction Model Using Decision Tree and Fuzzy Genetic enhances learning and understanding of computer network
Algorithm.," Procedia Technology, vol. 25, pp. 326-332, protocol concepts in middle east college," Airo International
2016. Research Journal, vol. 13, October 2017.
[22] Raheela Asif, Agathe Merceron, Syed Abbas Ali, and [25] J. Chandrasekaran, H. Feng, Y. Lei, D. R. Kuhn, and
Najmi Ghani Haider, "Analyzing undergraduate students' R. Kacker, "Applying Combinatorial Testing to Data Mining
performance using educational data mining," Computers & Algorithms," in 2017 IEEE International Conference on
Education, vol. 113, pp. 177-194, October 2017. Software Testing, Verification and Validation Workshops
[23] R. Hasan, S. I. Ali, and M. S. Hayat, "Enhancing (ICSTW), Tokyo, 2017, pp. 253-261.
Student’s Learning Experience at Middle East College by

~ 5 ~