Case Study of UPNM Students Performance Classification Algorithms

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/329528133
Case Study of UPNM Students Performance Classification Algorithms
Article in International Journal of Engineering and Technology · December 2018
DOI: 10.14419/ijet.v7i4.31.23382
CITATIONS
0
READS
139
3 authors, including:
Some of the authors of this publication are also working on these related projects:
Situational awareness for malaysian military observers View project
Keyword Patterns Analysis on Military Knowledge using Data Mining Technique for Military Personnel View project
Nur Diyana Kamarudin
National Defence University of Malaysia
12 PUBLICATIONS 15 CITATIONS
SEE PROFILE
Zuraini Zainol
National Defence University of Malaysia
46 PUBLICATIONS 106 CITATIONS
SEE PROFILE
All content following this page was uploaded by Zuraini Zainol on 10 December 2018.
The user has requested enhancement of the downloaded file.

Copyright © 2018 Authors. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricteduse, distribution, and reproduction in any medium, provided the original work is properly cited. International Journal of Engineering & Technology, 7 (4.31) (2018) 285-289 International Journal of Engineering & Technology Website: www.sciencepubco.com/index.php/IJET Research paper Case Study of UPNM Students Performance ClassificationAlgorithms Syarifah B. Rahayu1*, Nur D. Kamarudin2, Zuraini Zainol3 1Cyber Security Centre, National Defence University of Malaysia, Sungai Besi Camp 57000 Kuala Lumpur, Malaysia2Computer Science Department, Faculty of Science and Defence Technology, National Defence University of Malaysia, Sungai BesiCamp 57000 Kuala Lumpur, Malaysia*Corresponding author E-mail: [e-mail] Abstract Most students have a problem to keep track on their learning performance. Some lecturers with high teaching hours and burden of ad-ministration jobs may have difficulty to identify weak and low performance students. In this study, three classification tech niques areapplied on educational datasets to predict the students’ performance based on coursework assessments. Thus, this prediction results mayhelp lecturers and students to improve their teaching and learning process. The objective of study is to predict students’ performancebased on coursework assessments using classification algorithms. The selected classification algorithms applied in this study such as J4 8Decision Tree, Naïve Bayes and kNN. WEKA is used as an experimental tool. The selected algorithms are applied on a data of studentdatabase of Data Mining subject. Findings shows Naïve Bayes outperforms other classification algorithms with above 80% predictionrate. Thus, the students’ performance for Data Mining Subject is improved. As a conclusion, the classification algorithms can predictstudents’ performance on a particular subject based on coursework assessments. Keywords: Prediction; Comparative Analysis; Educational Data Mining 1. IntroductionEducational Data Mining (EDM) researches use data mining toolsto process large quantities of data to discover meaningful patternsin order to predict students’ performances to enhance teaching andlearning outcomes. These researches can also be used as a plat-form to alert student on the risk of failure and to provide recom-mendations for student improvement in their learning process.One of the criteria for a high quality university is based on itsexcellent record of academic achievement [1]. Therefore, studentperformance is a crucial part in higher learning institution. A stu-dent performance is often measured based on the subject workassessments and final exam. The proposed methodology is to ana-lyze students’ performance of a particular subject. The findingsare used for predicting their performance before they are taking afinal exam. Thus, it will assist the lecturers or educators to identifystudents who need supports to perform well in the final exam.Besides, students can improve their learning process in order topass the subject [2]. The objective of this study is to predict thestudents’ performance based on Malaysia Grading System. Theseperformances are predicted using three different classificationalgorithms, for example, J48 Decision Tree, Naïve Bayes andkNN.The rest of this paper is organized as follows. Section 2 presentsthe background and related work to this study. In Section 3, wedescribed the framework of our proposed research. Section 4 dis-cusses the experiment and results. Finally, we conclude this paperwith future work in section 5.2. Background and Related WorksIn this section, some related topics on data mining, knowledgediscovery in databases, classification algorithms and reviews onrelated work are discussed.2.1. Data Mining and Knowledge Discovery in DatabaseData Mining (DM) and Knowledge Discovery in Databases (KDD)are two terms that are often used interchangeably. KDD can bedefined as a process of finding useful information and patterns indata [3]. In KDD, DM is placed in the fourth steps of the KDDprocess. Technically, the KDD process consists of five main stepssuch as selection, pre-processing, transformation, data mining, andinterpretation or evaluation (see Fig 1).According to [3], DM is often applied to extract hidden informa-tion and useful patterns using algorithms from massive amounts ofdata which is derived by the KDD process. Such valuable infor-mation and patterns may assist the top level managers in decisionmaking. DM has been applied in various application areas such asmarket based analysis, healthcare [5], smart homes [6], business,text documents [7-10], environmental studies [11, 12], flood de-tection [13], crime investigation, fraud detection, geology, foodmicrobiology, astronomy, etc. Researchers [14] summarized somecommon data mining tasks and techniques (see Table 1). Thesetasks and techniques can be applied individually or they can becombined together to perform more sophisticated processes.

286 International Journal of Engineering & Technology Fig. 1: Knowledge Discovery in Databases (KDD) process adopted from [4] 2.2. Classification AlgorithmsClassification is a supervised learning where the classes are oftendetermined before data can be mined [3]. Technically, classifica-tion will assign the data into several predefined classes. Classifica-tion technique is often applied for predicting or describing datasetor nominal categories. Each classification technique (see Table 1)will apply a learning algorithm to identify a model which is bestfitted the relationship between the set of attributes and the classlabel (predefined class) of the input data. The model that has beenproduced by a learning algorithm should be able to fit the inputdata and predict the class label of the records correctly [15]. Table 1: Data Mining Tasks and Techniques Adopted From [14] DM Tasks DM Techniques Classification Decision Tree Induction, Bayesian Classifica-tion, Fuzzy Logic, Support Vector Machines(SVM), k-Nearest Neighbors (K-NN), RoughSet Approach, Genetic Algorithm (GA), etc. Clustering Partitioning Methods, Hierarchical Methods,Density-based Methods, Grid-based Methods,etc. Association Rules Frequent Item set Mining Methods (e.g.,Apriori, FP-Growth) Some examples of classification technique are detecting spamemail messages based on the message header and content, catego-rizing cells as malignant or benign based on the result of MRI,identifying credit risks based on bank loan, predicting students’performance, etc.In [16-19], the Decision Tree (J48), Bayesian Classifier and k-Nearest Neighbor (kNN) classifiers have been implemented toevaluate students’ performances based on several observationalattributes such as accumulated exam grades, percentages or clas-ses (i.e distinction, fail etc). Based on comparative analysis ofclassifier in [16], the Bayesian classifier outperformed the deci-sion tree and kNN classifier on predicting students’ performancesvia average True Positive (TP) rate. However, in analyzing the TPrate for each classes (Distinction, First, Second, Third and Fail); ithas been observed that, the prediction rates are not uniform amongclasses. Hence, the gap of prediction rate among classes is variedalmost 90% in some cases. This might be due to the insufficientdata of certain classes especially in distinction and fail classes.To discover the optimal classification model for decision tree,research in [20] did the comparison of different algorithms com-prises of J48, ID3, C4.5, REPTree, Random Tree and RandomForest. Out of six decision tree algorithms, the highest percentageis achieved using the model relying on the algorithm J48. Basedon 161 questionnaires, two researchers from University of Basrahhave analyzed and assisted academic achievers in higher educationusing Bayesian Classification Method [21]. For attribute selection,questions with high correlation averages have been adopted toenhance the accuracy of classification.Recent work has been done to demonstrate the efficiency of Semi-Supervised Learning (SSL) methods for the performance predic-tion of high school students using their final examination assess-ment percentage [22]. In this work, various SSL algorithms suchas Self-training, Co-training, Democratic Co-learning, Tri-training, De-Tri training and RASCO are implemented in KEELSoftware tool. In addition, Friedman Aligned Ranks nonparamet-ric test is used to measure the performances of these algorithms.Moreover, in second phase of experiments, the performance ofSSL classifiers have been compared with supervised method, Na-ïve Bayes. From the observation, it can be concluded that SSLalgorithm are comparatively better than the respective supervisedalgorithm, Naïve Bayes based on both measurement; the accuracyand Friedman Aligned Rank.A comprehensive survey is then carried out by the Indian Re-searchers to discuss about the current approaches and potentialareas in EDM [23]. This paper reported the details of researchesdone in the area of education in tabular form describing methodol-ogies and findings of each research and identifies potential re-search areas for future scope. Similar research is conducted by theresearcher in [24] where the new potential domains of EDM havebeen proposed. According to this paper, EDM data is not limitedto predict the student’s performance but can also be utilized inother domains of education sector (i.e. optimization of resourcesor human resource purposes).In comparison of correlation among pre and post enrollment fac-tors and employability using data mining tools, many of today’sgraduates are lacking interpersonal communication skills, creativeand critical thinking, problem solving, analytical skills, and teamwork [25]. It has been concluded that cognitive factors such as setof behaviors, skills and attitudes play a significant role in predic-tion of student’s marketability after graduation. Another workpresented by Research Group for Work, Organizational, and Per-sonnel Psychology, Department of Psychology, KU Leuven, Leu-ven, Belgium stated that employability is in strong correlationwith competences and dispositions [26].Motivated by the previous researches, this research attempts toevaluate the performances of several students by measuring theirsubject work assessment percentages (Quizzes, Tutorials and Test)via predetermined classes endorsed by the university and Malaysi-an Grading System to predict their performance in final exam. Tothe best of our knowledge, this is the only research paper thatdiscusses the student’s performance prediction in Malaysia basedon Malaysian Grading System apart from using the students’CGPA. We proposed this research as a preliminary assessmenttool where we narrower the scope of research to cater early recog-nition of student who needs help in certain subject not the wholeperformances of student from his or her CGPA. In addition, byanalyzing the distributed rank or weightage on each assessmentusing decision tree, this research will offer guidance to a lectureron improving the teaching plan based on the learning outcomes incertain assessment as well as to identify weak students to improvethe students’ learning process prior to the final exam. Future con-tribution will be the automatic application on a platform that isable to read, analyze and predict the outcome of student’s progressbased on certain assessments in difficult or challenging universitysubjects for intelligent tutoring or lecturing applications.3. Research FrameworkFigure 2 illustrates the proposed framework for predicting stu-dents’ performance in Data Mining subject. The first stage is datacollection. The data about students related to a particular subject