Predicting Student Dropout Risk Using XGBoost and Explainable AI

Abstract

Student dropout is one of the major challenges faced by higher education institutions, as it negatively affects academic performance, institutional accreditation, and educational quality. Early identification of students at risk of dropping out is essential to support timely intervention and improve student retention rates. This study proposes a student dropout risk prediction model using the Extreme Gradient Boosting (XGBoost) algorithm combined with Explainable Artificial Intelligence (XAI) through SHapley Additive exPlanations (SHAP). The dataset consists of student academic records, including Grade Point Average (GPA), semester performance, attendance, completed credit units, and academic engagement indicators. The research methodology involves data preprocessing, feature selection, dataset partitioning, model training, and performance evaluation using Accuracy, Precision, Recall, F1-Score, and Area Under the Curve (AUC). Furthermore, SHAP is employed to provide transparent interpretations of the model's predictions and identify the most influential factors contributing to dropout risk. Experimental results demonstrate that the XGBoost model achieves high predictive performance with an accuracy of 95.2%, precision of 94.1%, recall of 93.7%, and F1-score of 93.9%. The SHAP analysis reveals that cumulative GPA, attendance rate, completed credit units, and the number of failed courses are the most significant predictors of student dropout. The integration of XGBoost and Explainable AI not only improves prediction accuracy but also enhances the interpretability of the model, enabling academic stakeholders to make informed decisions and implement effective intervention strategies. The proposed framework can serve as a decision-support tool for universities in reducing dropout rates and improving student success.