Comparative Analytics and Predictive Modeling of Student Performance Through Data Mining Techniques

Abstract

As a matter of fact Data Mining is increasingly used as an emerging research domain in educational fields as well for extracting useful and previously unknown patterns from educational repositories. This research is an attempt to predicting the performance of the students for improving it in future through experimental analysis and predictive modeling. This experimental analysis and predictive modeling in learning process of students can be used as a better instrumental. In this study it was perceived that educational repositories/databases can have sufficient facts about learning attitude and capability etc. These factors are required to be understood for discovery of new hidden patterns and knowledge. These hidden patterns and knowledge can be later used for performance enhancement of the students through strategic planning and proper action by the university/college authorities. Researcher observed that there is lack of predictive studies and models available in Ethiopian context to accurately determine the influencing factors of the students’ performance in academics by categorizing student status in to drop out/fail, poor, good, excellent, or an average performer. Researcher also observed that there are many factors that affects student’ performance in university level education and some of selected factors in this research study are Gender, Department, Course, and Credit hours etc. Many educational institutions have still no enough strategic plan to predict or determine the student performance in order to improve it, reduce drop out and help to implement the curriculum / academic policies based on student performance and status. This research aims to conduct a comparative analysis and predictive modeling for knowing the student performance status through data mining techniques so as to improve their performance and status. This study used the KDD process model to find and interpret patterns in respositories. Decision tree (J48 and Random Forest), Bayes (NaiveBayes and BayesNet) and Rule Based (JRip and PART) algorithms are used for classification. The result of this study reveals that overall accuracy of the tested classifiers is above 80%. In addition, classification accuracy for the different classes reveals that the predictions are worst for fail class and fairly good for the average class. The J48 and JRip classifiers relatively produces highest classification accuracy for the average performer/ status. Finally the study suggests that data mining can be used as significant technique to figure out student performance based on salient factors affecting.

Keywords: Data Mining, Decision Tree, Prediction, Naive Bayes, Rule Based