Fernando Hose, Robet Robet, Hendri Hendri

Analisis Komparatif Model Regresi Machine Learning untuk Prediksi Prestasi Akademik Siswa dengan Optimasi Hyperparameter

Introduction

Analisis komparatif model regresi machine learning untuk prediksi prestasi akademik siswa dengan optimasi hyperparameter. Analisis komparatif model machine learning (CatBoost, LightGBM) untuk prediksi prestasi akademik siswa dengan optimasi hyperparameter. Membangun sistem peringatan dini yang efektif.

Abstract

Low accuracy in the early identification of at-risk students often hinders timely academic intervention. This study analyzes and compares seven machine learning algorithms to predict student academic achievement, aiming to provide a foundation for a reliable early warning model. The dataset includes 2.392 students with 15 features covering demographics, learning behavior, and environmental support. Model training was performed using GridSearchCV optimization combined with stratified cross-validation to mitigate overfitting.Performance was evaluated using MAE, RMSE, and R². The results show CatBoost performed the best R² = 0,774; RMSE = 0,581; MAE = 0,306) followed by LightGBM (R² = 0,771) and Gradient Boosting (R² = 0,767), while MLP showed the lowest performance. Feature importance analysis placed GPA as the dominant predictor, followed by absenteeism and weekly study time. These findings affirm the superiority of boosting-based models in capturing complex nonlinear relationships and provide a practical framework for educational institutions to build data-driven early warning systems.

Review

This study addresses the critical challenge of accurately identifying at-risk students early to enable timely academic interventions, a highly pertinent area in educational research. The authors effectively compare seven machine learning algorithms, employing robust methodologies such as GridSearchCV optimization combined with stratified cross-validation to enhance model reliability and mitigate overfitting. The extensive dataset of 2,392 students with 15 diverse features provides a solid foundation for the analysis. The clear reporting of performance metrics (MAE, RMSE, R²) consistently highlights the superior performance of boosting-based models, particularly CatBoost, LightGBM, and Gradient Boosting, over MLP. Furthermore, the identification of GPA, absenteeism, and weekly study time as dominant predictors offers actionable insights for educators seeking to establish data-driven early warning systems. While the study presents compelling findings, there are several areas that could enhance its clarity and impact. The abstract could benefit from explicitly defining the "academic achievement" target variable (e.g., end-of-year GPA, specific course score) to contextualize the regression metrics. A brief mention of the specific educational level or institution from which the data was collected would also improve generalizability and context. Additionally, while "demographics, learning behavior, and environmental support" are mentioned, listing a few representative features from each category would give readers a clearer picture of the input variables. Finally, a brief discussion on the practical implications of the computational cost of the best-performing models, especially for real-time early warning systems, could be valuable. Overall, this research represents a valuable contribution to the field of educational analytics and machine learning applications in education. The methodological rigor, comparative analysis, and practical insights regarding boosting models and key predictors make a strong case for its utility in developing effective early intervention strategies. The study successfully affirms the capability of advanced machine learning techniques to capture complex relationships in educational data, providing a tangible framework for institutions. With minor clarifications, this paper offers significant practical and theoretical implications and is highly recommended for publication.

Full Text

Comments

You need to be logged in to post a comment.