Multi-Level Student Performance Prediction Using Machine Learning Based on Socio-Demographic and Development Indicators
Main Article Content
This study investigates the application of machine learning techniques to predict student performance across multiple educational levels using socio-demographic and development indicators. A combined dataset covering Primary, Lower Secondary, and Upper Secondary levels was utilized to develop a multi-level predictive framework. A clean machine learning pipeline was implemented, including the removal of potential data leakage features, feature engineering through development index encoding, and standardized preprocessing. Two ensemble models, Random Forest and XGBoost, were evaluated using Mean Absolute Error, Root Mean Squared Error, and the coefficient of determination. The results show that Random Forest outperforms XGBoost, achieving the highest predictive performance with an R² score of 0.716, indicating strong explanatory power. Feature importance analysis reveals that education level is the most influential predictor, followed by temporal and macro-level variables such as year, regional classification, and development index, highlighting the significant role of socio-economic context in shaping student outcomes. Further analysis indicates that prediction performance varies across educational stages and that the model tends to underestimate high-performing students and overestimate low-performing students, suggesting limitations in capturing extreme values. Overall, the findings demonstrate that integrating machine learning with multi-level and macro-level analysis provides meaningful insights into student performance and supports data-driven educational policy and planning.