An Evaluation of Ensemble and Non-Ensemble Data Mining Techniques in Predicting Students' Performance on E-learning Dataset

The excessive use of e-learning technology today has resulted in a massive growth in educational data. Students' interactions with the e-learning environment, particularly learning management systems, create huge amount of data within the shortest period of time. The data contains hidden information about students' engagement in various e-learning activities, which can be linked to their performance. Predicting student performance based on the usage of e-learning systems in educational institutions is a big concern, and it has become critical for educational administrators to better understand why so many students do badly or fail their courses. However, due to the numerous features that influence their performance, making a prognosis is challenging. This study aims to compare various ensemble techniques against their nonensemble counterpart in predicting students’ performance on data generated from learning management system. Five popular algorithms were used: Decision Tree (DT), KNearest Neighbor (KNN), Discriminant Analysis (DA), Nave Bayes (NB) and Support Vector Machine (SVM). To improve the performance of the classifiers, ensemble techniques such as RUSBoost, Bag and AdaBoost were employed to increase the accuracy of the students' performance prediction models. The findings revealed that after applying ensemble methods, it achieved a higher accuracy was obtaine