Predicting Student Performance Using Ensemble Learning: An Xgboost Approach with Model Evaluation and Insight

Abstract:

Predicting student academic performance was important for identifying students at risk of poor achievement and supporting timely educational interventions. This study developed and evaluated an XGBoost-based machine learning model for predicting student performance using academic, behavioural, and socioeconomic factors. The study used a dataset obtained from the Kaggle repository containing 6,607 student records and 20 features. The data were preprocessed to address missing values and prepare categorical and numerical variables for modelling. An XGBoost regression model was trained using 80% of the dataset and evaluated on the remaining 20% using Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R²). The model achieved an MSE of 4.75, an MAE of 0.97, and an R² of 0.66, indicating that it explained approximately 66% of the variation in student performance. Feature-importance analysis showed that hours studied, previous scores, and attendance were the most influential predictors, while sleep, tutoring, parental involvement, and socioeconomic factors also contributed to prediction. However, the model showed greater difficulty predicting students at the higher end of the performance distribution. The findings demonstrated the usefulness of XGBoost for student performance prediction while highlighting the need for richer features, balanced datasets, and further model optimization.