A data mining approach to predict academic performance of students using ensemble techniques

Abstract

Recently, Educational Data Mining (EDM), emerged as a new area of research due to the enlargement of various statistical methods used to explore data in educational settings. One of the applications of EDM is the prediction of student performance. The application of Data Mining methods in an educational setting is able to discover some hidden knowledge and patterns which will help in decision making for administrators for enhancing the educational system. In a web based education system, the behavioral features of learners is very significant in showing the interaction between students and the LMS. In this paper, our aim is to propose a new performance prediction model for students which is based on data mining methods which includes new features known as behavioral features of students. The proposed predictive model is evaluated using classifiers like Naïve Bayesian (NB), Decision Tree (DT), K-Nearest Neighbor (KNN), Discriminant Analysis (Disc) and Pairwise Coupling (PWC). Additionally, so as to enhance the classifiers performance, the ensemble methods such as AdaBoost, Bag and RUSBoost were used to enhance the accuracy of the performance model of the students. The achieved results shows that there exist a strong relationship between behavior of students and their academic performance. The accuracy of the proposed model achieved 84.2% with behavioral features while it achieved 72.6% without behavioral features. More so, an accuracy of 94.1% was gotten when the ensemble methods were applied to the classifiers to improve the academic performance. Therefore the result gotten shows the reliability of the proposed model.