Enhancing Graduate Education Assessment: A Machine Learning Approach to GPA Prediction for Medical Students

Background In recent years, Chinese medical postgraduate education has undergone significant transformation, with enrollment soaring to 156,000 students in 2023, accounting for 12% of the nation’s total postgraduate admissions. Recognizing high-achieving students at an early stage and learning from their success can change the way to influence future educational dilemmas. While the existing evaluation system remained limited in its ability to prospectively predict academic performance variability. This critical gap underscores the need for innovative, data-driven approaches to transcend conventional assessment paradigms. Our study used machine learning techniques to predict the academic outcomes (GPA) of medical postgraduate students in a certain university, providing evidence-based strategies that could be used to improve educational practices and improve student performance. Methods In this study, we worked with 1,133 postgraduate students at Southern Medical University (2020 cohort) while analyzing 42 variables, including demographic, undergraduate performance, postgraduate transition variables, and measures of their self-assessment. Using the Boruta algorithm, we identified the most important predicting features and then tested eight learning machine models to find the best one. Furthermore, we applied SHAP (Shapley Additive Explanations) to derive interpretable insights into the most critical features of success. Finally, we submitted this work as an interactive web application that allowed academic leaders to predict their students’ GPAs and provide much-needed proactive support. Results XGBoost model crushed the competitions, delivering higher predictions (AUC = 0.744, Accuracy = 82.8%, F1 = 0.902). SHAP analysis exposed the secret formula for success. A student’s career ambition, undergraduate ranking, and core subject mastery weren’t just important, but they were also game changers. Based on the results of our study, we built a smart web tool that could turn data into action, giving educators a crystal ball to flag high-potential students early, personalize support proactively, and allocate resources smartly. Our results were not just number crunching; it will also be a new playbook for medical postgraduate students’ education. Conclusion This research showed that XGBoost model did not just predict academic performance; it also revealed hidden pathways for students’ success. We found that students should focus on ambition, track record, and their mastery of core subjects to drive their success based on machine learning and SHAP models. More importantly, we developed a clever, user-friendly tool that might help educators identify students with high potential earlier and intervene with support ahead of challenges.