A Multiobjective, Stable, and Fair Feature Selection and Ensemble Learning Framework for Explainable Student Academic Performance Prediction

Early intervention and data-driven educational decision-making depend on accurate student academic performance prediction; unfortunately, current methods frequently have high dimensionality, poor interpretability, and inadequate resilience. To obtain precise, understandable, stable, and equitable academic performance prediction, this study suggests an improved framework that combines a multiobjective Rüppell’s fox optimizer (RFO) with a gradient boosting classifier (GBC). An explicit trade-off between predicted accuracy and interpretability is made possible by the formulation of the feature selection issue as a multiobjective optimization assignment that simultaneously minimizes classification error and feature subset size. The proposed RFO-based method maintains good predictive performance while reducing the feature space to informative features, dependent on the chosen trade-off parameter, using the students’ academic performance dataset. The most effective RFO–GBC design outperforms traditional classifiers with an overall classification accuracy of 91.46% and a weighted F1-score of 91.30%. The efficacy of the proposed multiobjective formulation is demonstrated by a Pareto analysis, which shows that competitive accuracy (90.87%) can be achieved using a reduced feature subset. A stability analysis verifies the resilience of feature selection by several separate runs. The most significant predictors are completed credits and prior academic accomplishment, according to SHAP-based explainability, which further improves the interpretability of the model. Furthermore, little predictive bias across sensitive attributes is indicated by a fairness evaluation based on equal opportunity and demographic parity, which supports ethical deployment in educational situations. Overall, the proposed framework offers a robust, interpretable, and fairness-aware solution for student academic performance prediction, contributing to trustworthy educational data mining.