Scholarlink

<p>This cross-sectional study aimed to identify key predictors of academic performance and develop robust machine learning models for predicting cumulative grade point average (CGPA) among 3,639 undergraduate students at Universiti Brunei Darussalam. Data on demographic characteristics, mental well-being, chronotype, physical activity, sedentary behavior, burnout, learning styles, motivation, personality traits, self-efficacy, and grit were collected using an online survey. Advanced feature selection techniques and various class imbalance handling methods were employed. Machine learning algorithms, including Random Forest, XGBoost, Logistic Regression, and a Stacking Ensemble, were evaluated. The XGBoost model with oversampling emerged as the best-performing model (F1-score: 73.1%, AUC-PR: 71.3%). Feature importance analysis using SHapley Additive exPlanations (SHAP) revealed that age, mental well-being, extraversion, and conscientiousness were the most influential predictors of CGPA. The findings offer actionable insights for designing targeted interventions and support systems to optimize academic success, while the developed prediction pipeline enables proactive identification of at-risk students. This study contributes to educational research by integrating machine learning with a comprehensive set of predictors, providing a holistic approach to understanding and predicting academic performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting academic performance through machine learning: integrating demographic, psychological, and behavioral predictors using explainable AI

  • 期刊论文

  • 发表日期：

    2026年5月16日

查看全文
  • Wen Jun Lau,
  • Hanif Abdul Rahman

摘要

This cross-sectional study aimed to identify key predictors of academic performance and develop robust machine learning models for predicting cumulative grade point average (CGPA) among 3,639 undergraduate students at Universiti Brunei Darussalam. Data on demographic characteristics, mental well-being, chronotype, physical activity, sedentary behavior, burnout, learning styles, motivation, personality traits, self-efficacy, and grit were collected using an online survey. Advanced feature selection techniques and various class imbalance handling methods were employed. Machine learning algorithms, including Random Forest, XGBoost, Logistic Regression, and a Stacking Ensemble, were evaluated. The XGBoost model with oversampling emerged as the best-performing model (F1-score: 73.1%, AUC-PR: 71.3%). Feature importance analysis using SHapley Additive exPlanations (SHAP) revealed that age, mental well-being, extraversion, and conscientiousness were the most influential predictors of CGPA. The findings offer actionable insights for designing targeted interventions and support systems to optimize academic success, while the developed prediction pipeline enables proactive identification of at-risk students. This study contributes to educational research by integrating machine learning with a comprehensive set of predictors, providing a holistic approach to understanding and predicting academic performance.