DEVELOPMENT OF A PREDICTIVE MODEL FOR STUDENT PERFORMANCE CLASSIFICATION USING ENSEMBLE LEARNING

This study presents the development of a predictive model for the classification of student academic performance using ensemble learning techniques. Data containing demographic information, UTME results, SSCE grades, and 100-level CGPA records were collected from the Department of Computer Science. Following preprocessing, categorical variables were transformed into numerical representations, and relevant features were identified using the mutual information method. Two ensemble algorithms—Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM)—were formulated and evaluated using five simulation runs of varying train-test splits. Model performance was assessed using accuracy, precision, recall, and F1‑score. Results showed that 100‑level CGPA values were the most influential predictors of final academic performance, while demographic features contributed minimally. Across the simulations, both ensemble models demonstrated strong predictive capabilities, with LightGBM consistently achieving slightly higher accuracy and better generalization. The study concludes that ensemble learning is an effective approach for classifying student performance and offers valuable insights for early intervention, academic advising, and institutional decision-making. Recommendations include expanding the dataset across multiple departments and exploring additional ensemble techniques to further enhance prediction robustness.