SHAP Explainability: An Ensemble Learning Approach for Student Performance Prediction
Ghalia Nassreddine, Loubna Saleh, Mohamad Al Majzoub, Amal El Arid · 2025
Student academic performance is a crucial metric of educational advancement, influenced by factors such as gender, age, instructional staff, and learning environments. Forecasting and evaluating student performance are essential for enabling educators to identify deficiencies and facilitate grade enhancement. A comprehensive understanding of feature contributions is essential for enhancing model interpretability and enabling informed decision-making in academic institutions. Explainable AI (XAI) is a set of techniques and strategies that aim to provide transparent and comprehensible justifications for the decisions made by artificial intelligence and machine learning models. This paper proposes an XGBoost ensemble learning model with the ADASYN technique to tackle the problem of the imbalanced dataset. A hyperparameter tuning phase was conducted subsequent to stratified k-fold cross-validation to enhance XGBoost performance. The efficacy of the suggested method was measured by accuracy, precision, and recall in identifying various performance levels. The results of applying this approach to a dataset collected from Kaggle demonstrate its effectiveness and robustness. Next, the significance of features was evaluated using the SHAP methodology, calculating each feature’s contribution to the model’s predictions. The findings highlight the importance of student attendance, as increased absences negatively impact performance. Extracurricular involvement may adversely impact student grades. Furthermore, the duration of study has a beneficial effect on academic success.