An explainable AI-based approach for predicting undergraduate students academic performance

Abstract

The accurate prediction of students' academic achievement has garnered considerable attention in the research community due to its importance in understanding students' progress and assisting them in achieving success. This study presents a novel approach for predicting undergraduate student's performance in the context of Bangladesh. The dataset contains 872 student records from multiple institutions. Initially the dataset was produced utilizing data-preprocessing techniques such as one-hot encoding, column remaining, and managing missing values. SMOTE (Synthetic Minority Oversampling Technique) and normalizing algorithms were employed to attain data balance and feature scaling, respectively. Afterwards, a total of seven distinct machine learning (ML) classifiers, with hyperparameter tuning, were employed to train and test in order to achieve the prediction of students' academic performance. Furthermore, a custom stacking ensemble classifier was utilized, which attained an accuracy of 86.38 %. This classifier outperformed the machine learning classifiers based on the four performance evaluation metrics. Two eXplainable Artificial Intelligence (XAI) algorithms, namely SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations), were integrated to provide a comprehensible prediction of the best model and determine the significant factors. This approach provided transparency, fairness and reliability on prediction that improved student performance in the classroom and anticipation.