<![CDATA[This study focuses on predicting student academic performance while emphasizing model interpretability through Explainable Artificial Intelligence (XAI). The main objective is to identify potential academic risks using machine learning models and provide transparent explanations for their decisions. Historical student academic data were used to train and evaluate two classification models: Random Forest and XGBoost. The results show that both models exhibit strong predictive performance. Random Forest achieved an accuracy of 90.77% and a precision of 0.7500 for the risk class, while XGBoost attained a higher recall of 0.7000 with an accuracy of 89.23% and a precision of 0.6364. Both models achieved an identical F1-score of 0.6667 for the risk class. The application of XAI methods, namely SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations), revealed the main features influencing the predictions. Globally, G2 (previous period’s final grade), failures (number of failed courses), and absences were identified as the most critical factors. Local interpretations from SHAP and LIME also clarified individual predictions, both correct and misclassified. The study contributes to developing an accurate and transparent decision-support system to enable more personalized, effective, and data-driven academic interventions.]]>
Copyrights © 2026
