Predicting Student Withdrawal from UAE CHEDS Repository using Data Mining Methodology

Abstract

Early prediction of a student who is at risk of course dropout leads to student retention in the study course. The percentage of student dropout in higher education sector is high, and affects the students’ careers negatively and the institute’s program continuation. The purpose of this study is to predict and identify students who are likely to withdraw from an institute. This identification assists the institute’s advisor to take precautionary measures to retain this group of students. Also, the study aims to find the variable that is most efficient to lead to student dropout prediction. To fulfil the study’s aim, CRISP method was followed after reviewing research papers. A dataset of 1272 students’ data in size from Central Higher Education Data Store (CHEDS) has been fetched from Dubai’s governmental higher education institute. The demography of students is international background. Several model classifiers from Standard and ensemble were implemented to find the best answer to the research questions. Receiver Operator Characteristic (ROC) based on Area Under Curve (AUC) was used to assess the result plus other metrics. Research outcome, results showed that students who had low GPA, average register credit hours and fluctuating student’s enrollment status were more likely to withdraw from study course. Random Forest classifiers demonstrated the highest performance in prediction, and scored 87.8% in AUC with an accuracy of 84.82%. GPA and average register credit hours attributes were the most effective contributor in prediction.