A predictive approach based on machine learning for university student re-enrollment

The primary objective of universities is to ensure the academic success of their students. At Sidi Mohamed Ben Abdellah University (USMBA) in Fez, a study developed a predictive model to assess the risks of non-reenrollment using academic and socio-demographic data such as gender, place of birth, type of high school diploma, and program of study. To tackle this issue, machine learning algorithms like XGBoost, CatBoost, Balanced Random Forest, and EasyEnsembleClassifier were implemented and combined through a StackingClassifier, with the SMOTETomek method addressing class imbalance between reenrolled and non-reenrolled students. The model achieved an accuracy of 77%, excelling in predicting reenrollment (precision: 82%, recall: 92%), though its performance in detecting non-reenrollment was constrained by the diversity within this group. These results underscore the utility of ensemble techniques and resampling methods in identifying at-risk students, offering a solid foundation for dropout prevention and university retention strategies.