A predictive Modeling Framework for Identifying Key Factors Influencing Students’ Academic Performance in Secondary Schools Using Machine Learning

Abstract

This study develops a predictive modeling framework to identify key factors influencing student academic performance in public secondary schools. Using a dataset comprising socio-economic, demographic, and academic variables, three machine learning algorithms like Linear Regression, Random Forest, and XGBoost were evaluated. Feature selection was conducted using Linear Regression coefficients, Random Forest importance, and XGBoost importance to extract the most impactful predictors. The models were assessed using Root Mean Squared Error (RMSE) and the coefficient of determination (R²). Results indicate that the XGBoost feature selection combined with Linear Regression yielded the highest performance (RMSE = 40.182, R² = 0.331), demonstrating improved predictive accuracy compared to other combinations. The findings highlight the significance of factors such as study hours, attendance rate, teacher quality, and assignment completion in determining student outcomes. This research contributes to data-driven educational decision-making, enabling stakeholders to target interventions more effectively. Recommendations for policy, practice, and future research are also provided