Optimizing XGBoost with Optuna for Attendance-Based Prediction of Student Academic Success

Abstract

The growth of data analytics in higher education has increased the need for predictive models that identify students’ academic potential and support data-driven decision-making. Academic institutions are able to enhance student outcomes by implementing suitable interventions that are based on an accurate early prediction of Grade Point Average (GPA). This study aims to develop a practical and accurate GPA prediction model based on Extreme Gradient Boosting (XGBoost), enhanced through Optuna-based hyperparameter tuning, to support academic monitoring systems in higher education. This model is intended to assist academic monitoring systems in higher education. Using academic performance data, attendance records, and course load information obtained from 961 undergraduate students, a quantitative predictive modeling approach was implemented. The CRISP-DM framework was implemented during the modelling process, which included the following stages: data understanding, data preparation, modelling, evaluation, and deployment. To ensure the stability and relevance of the model, exploratory analysis and correlation assessments were implemented to determine feature selection. Optuna was employed to optimize hyperparameters, utilizing Bayesian optimization with adaptive trial pruning to efficiently examine the parameter space. Experimental results demonstrate that the Optuna-tuned XGBoost model achieved superior predictive performance compared to baseline XGBoost models and models optimized using Grid Search and Random Search. The proposed model attained a coefficient of determination (R²) of 0.8637 and a Root Mean Square Error (RMSE) of 0.1165, indicating improved accuracy and robustness in handling large prediction errors. To enhance practical applicability, the final model was deployed in a Streamlit-based web application that enables real-time GPA prediction and supports academic advisors. Overall, the findings confirm that Optuna-based hyperparameter tuning significantly enhances XGBoost performance and provides a solution for data-driven academic monitoring in higher education institutions.