On the Fairness of Ensemble Learning Methods in Student Dropout Prediction

Abdelghafour Aboukacem, Loubna Mekouar, El Houcine Bergou, Youssef Iraqi, Ismail Berrada

Abstract: Student Dropout Prediction (SDP) is a crucial task in Educational Data Mining (EDM). It aims to help institutions intervene early to improve retention. Despite notable advancements in the predictive performance of the developed models, the topic of fairness is usually overlooked. Algorithmic bias can lead to disparate impacts across demographic groups. In this study, we explore the use of ensemble learning, specifically bagging, boosting, voting, and stacking techniques to improve both the predictive performance as well as the fairness aspect. Using a real-world dataset that spans the whole K-12 system of Morocco with 14 sub-datasets, we evaluate various Machine Learning (ML) models trained using the previously mentioned ensemble learning techniques. Our study examines model fairness for several protected attributes, including gender, handicap, financial aid, and boarding school availability.

External IDs:doi:10.1007/978-3-031-98462-4_46