Learning Academic Performance Patterns for Explainable Educational Decision Support Systems

The early identification of risks and reliable predictive methods is important, This study examined academic performance patterns using an explainable machine-learning framework designed to support educational decision-making. A secondary dataset comprising 480 student records and 17 demographic, academic, behavioural, parental, attendance, and performance variables was analysed. After duplicate screening, 478 unique observations were retained. Academic performance was represented by three categories: Low, Middle, and High. Exploratory analysis, statistical association testing, and supervised machine-learning classification were conducted using multinomial logistic regression, decision tree, random forest, support vector machine, and XGBoost models. Stratified five-fold cross-validation was used for model evaluation. Random Forest demonstrated the strongest overall performance, achieving an accuracy of 0.778, macro-F1 of 0.786, and multiclass ROC-AUC of 0.913. Attendance, learning-resource visits, raised-hand participation, parent-related factors, and announcement viewing emerged as important predictive features. SHAP-based explainability showed that student absence days contributed most strongly to model predictions, followed by resource use and classroom engagement variables. The findings indicate that combining multiclass prediction with explainable artificial intelligence can provide interpretable academic-risk profiles rather than relying only on performance labels. The proposed framework may support teachers and educational administrators in identifying students who may require targeted academic support while preserving human oversight in intervention decisions.

Hali tarjima qilinmagan