Applying classification techniques to predict students' academic results

Abstract

The ability to predict students' academic results is due to the application of various models developed using Data Mining techniques. It is one of the many applications available in Educational Data Mining (EDM). Although there are many classification algorithms available and there have been many researches done to compare certain algorithms, this study aims to compare Bayes-based, ANN-based, Regression-based, SVM-based, Instance-based, Tree-based and Rule-based classification algorithms using a dataset downloaded online from UCI Machine Learning Repository. The data were preprocessed and categorized to different datasets before applying Data Mining techniques. Dimension reduction and data re-balancing were applied and proved to increase prediction accuracy. The algorithms were compared in terms of accuracy rate, precision rate, AUC and model building time. Tree-based algorithms performed better than other categories of algorithms. Within Tree-based algorithms, RandomForest algorithm proved to be the best classification algorithm. Tree-based and Rule-based algorithms reveal the key attributes contributing to accurate prediction of students' results.