An Empirical Study to Improve Multiclass Classification Using Hybrid Ensemble Approach for Students’ Performance Prediction

Improving machine learning algorithms has been the interest of data scientists and researchers for the past few years. Among the performance problems raised is the classification imbalance issues listed as the top ten. The present study makes comparison and analysis of 5 state-of-art classifiers, 5 ensembles classifiers and 10 resampling techniques for data imbalance. This is done via the used 4413 instances consisting of demographic, economic, and behavioural data from student information systems and e-learning, as well as engineering faculty for the first semester 2017/2018. The use of three sampling types was adapted for the analysis: oversampling, undersampling and hybrid. The experimental results prove to model students’ behaviour from e-learning data and bagging decision tree ensemble classifier produces the highest results. Lastly, a hybrid resampling technique, SMOTEENN consistently shows the top result compared to other resampling techniques.