Decision Trees for Very Early Prediction of Student's Achievement

Abstract

The prediction of students' academic achievement is crucial to be conducted in a university for early detection of students at risk. This paper aims to present data mining models using classification methods based on Decision Trees (DT) algorithms to predict students' academic achievement after preparatory year, and to identify the algorithm that yields best performance. The students' academic achievement is defined as High, Average, or Below Average based on graduation CGPA. Three classifiers (J48, Random Tree and REPTree) are applied on a newly created dataset consisting of 339 students and 15 features, at the College of Computer Science and Information Technology (CCSIT). The outcome showed the J48 algorithm had an overall superior performance compared to other algorithms. Feature selection algorithms were used to reduce the feature vectors from 15 to 4 resulting in improvements in performance and computational efficiency. Finally, the results obtained help to pinpoint the preparatory year courses that impact graduation CGPA. Timely warnings, and preemptive counseling towards improving academic achievement is possible now.