AI Early-Warning System for Student Dropout Risk — Mohammed VI Polytechnic University and Morocco's Ministry of National Education

Morocco's Ministry of National Education, Preschool and Sports partnered with Mohammed VI Polytechnic University (UM6P) to address student dropout using a large administrative dataset covering roughly 1.4 million student records across five academic years in the Fès-Meknès region.

The project aimed to build an explainable machine-learning system that could flag students at risk of dropping out early enough for schools to intervene, using SHAP-based interpretability so predictions could be understood by administrators rather than treated as a black box.

Researchers benchmarked five algorithms (Decision Tree, Random Forest, XGBoost, LightGBM and an ensemble voting approach) trained on more than 100 student-level features, then applied SHAP to surface which factors drove each risk prediction.

With a random data split, the best configuration reached 88% accuracy, 88% recall, 86% precision and an 87% AUC; the single best model (LightGBM on Level 7 students) reached 83% accuracy and 89% AUC. Performance dropped substantially when evaluated with a more realistic chronological (year-over-year) split, which the authors partly attribute to COVID-19-era disruption of typical dropout patterns.

The authors present the system as a promising but preliminary research result confined to one region, cautioning that the headline accuracy figures may not hold in real deployment and noting that the dropout class's small share of records complicates classification.