Prediction of Medical Students' Performance in Summative Assessment Using Machine Learning Algorithms

Early identification of medical students at risk of poor academic performance is essential for timely academic intervention and student support. This study aimed to develop and evaluate machine learning models to predict the performance of second-year MBBS students in summative assessments.

This observational analytical study was conducted at All India Institute of Medical Sciences, Kalyani, using academic data from five consecutive MBBS batches (2019-2023). Variables included demographic information, attendance records, internal assessment scores, previous academic performance, and final summative examination results in microbiology, pathology, and pharmacology. Multiple machine learning algorithms including logistic regression, decision tree, random forest, support vector machine (SVM), k-nearest neighbors, Naïve Bayes, XGBoost, and LightGBM were implemented using R statistical software. The models were trained using data from four batches and validated using the fifth batch. Performance was evaluated using accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and area under the receiver operating characteristic curve (AUC).

Machine learning models demonstrated strong predictive performance across all three subjects. Among the algorithms evaluated, the SVM model showed the highest sensitivity (89%) in identifying students at risk of failing in final examinations. Specificity across models reached 100%, indicating that students predicted to pass were accurately classified. Random forest, XGBoost, and LightGBM also demonstrated high predictive accuracy and AUC values.

Machine learning-based predictive models can effectively identify medical students at risk of poor academic performance. Integrating predictive analytics into medical education systems may enable early academic support, reduce student stress, and improve educational outcomes.