Predictive Modeling and Explainability for Academic Dropout Risk Detection Using Machine Learning

Student dropout is a major challenge for higher education institutions, especially in developing countries where socioeconomic disparities and resource constraints worsen the issue. This study proposes a framework using machine learning (ML) and explainable AI (XAI) tools to predict and understand dropout risks in undergraduate students. An ensemble model combining XGBoost, LightGBM, and neural networks achieved state-of-the-art performance (AUC-ROC: 0.995, F1-score: 0.875, Recall: 83.33%). Key predictors include academic progress, socioeconomic indicators, and behavioral metrics, with SHAP and LIME offering actionable insights for targeted interventions. The model balances precision and recall, ensuring accurate identification of at-risk students while maintaining interpretability. Additionally, the incorporation of explainable AI techniques empowers academic administrators to trust and act upon model predictions, enabling the design of targeted strategies to improve student retention and institutional performance. Scalable and resource-efficient, the framework is well-suited for public institutions with constrained resources. The model was developed and validated using data from a public university, taking into account institutional constraints such as academic modality, course load structures, session availability, and student socioeconomic profiles.