Abstract
BACKGROUND: Accurately predicting academic performance among medical postgraduate students is crucial for understanding educational outcomes and providing effective early academic guidance. Traditional statistical approaches often struggle to balance predictive performance with interpretability, particularly when handling complex relationships among academic and psychosocial factors. METHODS: A semi-structured survey was administered to medical postgraduate students at a Chinese medical university, yielding a final sample of 1,091 participants. GPA was dichotomized into two categories: outstanding academic performance (GPA ≥ 80) and non-outstanding academic performance (GPA < 80). Feature selection was performed using the Boruta algorithm. Logistic regression and XGBoost models were developed and evaluated on a held-out test set. Model performance was assessed using the area under the receiver operating characteristic curve, accuracy, and complementary validation metrics. Shapley Additive Explanations (SHAP) analysis was applied to interpret the contributions of key predictors. RESULTS: Both models demonstrated acceptable predictive performance. Undergraduate academic achievement emerged as the most influential predictor of GPA classification, followed by selected psychosocial characteristics and foundational academic skills. Shapley Additive Explanations (SHAP) interpretation provided transparent insights into the relative importance and directionality of these predictors. CONCLUSION: This study presents an interpretable machine learning framework for predicting academic performance in medical postgraduate education. By combining predictive modeling with explainable techniques, the proposed approach supports reliable performance assessment while maintaining transparency, offering a methodological foundation for future research and cautious application in educational analytics.