Predictive Analysis of Bangladeshi Students' Academic Performances Using Ensemble Machine Learning with Explainable AI Techniques

Educational institutions benefit immensely from the ability to predict and address students' academic performance proactively, such as timely intervention. Despite some prior studies on Bangladeshi students, there is a need to expand the study on students' performance prediction with a focus on secondary and higher-secondary students. In this research, we want to fill this gap by analyzing the factors influencing the academic journey of Secondary School Certificate (SSC) and Higher Secondary Certificate (HSC) students in Bangladesh. Our research dataset was collected from a sizeable number of SSC and HSC students. It was designed to encompass a spectrum of relevant factors within a student's life, recognizing that SSC and HSC examinations mark critical milestones in their educational journey. Our primary objective is to predict students' performances while concurrently identifying factors shaping their academic trajectory. To achieve this, we first employed a range of individual machine-learning models with sophisticated hyperparameter tuning strategies. Then, we introduced a novel stacking ensemble approach on top of those baseline models. A significant insight from our research underscores the pivotal role of feature scaling in prediction accuracy. Our ensemble model surpassed the performance of base models, achieving a remarkable 82.23% accuracy for the SSC results and an even more impressive 86.89% for the HSC results. Subsequently, we utilized the SHAP (SHapley Additive exPlanations) method to give insights into the prediction results so as to improve the confidence in the prediction model.