Scholarlink

<p>The development of digital learning environments has generated rich educational data capable of supporting early prediction of student outcomes. In this study, seven diverse datasets, spanning demographics, parental education, assessment history, and VLE engagement, were integrated into a unified machine-learning pipeline. Four supervised algorithms (Logistic Regression, Random Forest, XGBoost, and Multi-Layer Perceptron) were trained using SMOTE and stratified fivefold cross-validation. XGBoost demonstrates the strongest performance among the evaluated models, achieving 95.04% accuracy, ROC-AUC of 0.9879, F1-score of 0.95, precision of 0.96, and recall of 0.95 on the test set. Cross-validation further confirmed the model’s robustness (mean ROC–AUC = 0.9879). The Bagging ensemble also demonstrated competitive performance, achieving a cross-validation accuracy of 0.9487, a cross-validation F1-score of 0.9470, and a test accuracy of 0.9504. To ensure interpretability, SHAP and LIME were employed, revealing cumulative assessment performance, withdrawal patterns, and engagement intensity as the most influential predictors. The results demonstrate that combining strong predictive algorithms with explainable AI (XA) creates a reliable early-warning system capable of supporting targeted academic interventions and improving institutional decision-making.</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

Ensemble machine learning framework with SHAP and LIME for accurate early prediction of student success in online learning environments

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月28日

查看全文
  • Essa E. Almazroei

摘要

The development of digital learning environments has generated rich educational data capable of supporting early prediction of student outcomes. In this study, seven diverse datasets, spanning demographics, parental education, assessment history, and VLE engagement, were integrated into a unified machine-learning pipeline. Four supervised algorithms (Logistic Regression, Random Forest, XGBoost, and Multi-Layer Perceptron) were trained using SMOTE and stratified fivefold cross-validation. XGBoost demonstrates the strongest performance among the evaluated models, achieving 95.04% accuracy, ROC-AUC of 0.9879, F1-score of 0.95, precision of 0.96, and recall of 0.95 on the test set. Cross-validation further confirmed the model’s robustness (mean ROC–AUC = 0.9879). The Bagging ensemble also demonstrated competitive performance, achieving a cross-validation accuracy of 0.9487, a cross-validation F1-score of 0.9470, and a test accuracy of 0.9504. To ensure interpretability, SHAP and LIME were employed, revealing cumulative assessment performance, withdrawal patterns, and engagement intensity as the most influential predictors. The results demonstrate that combining strong predictive algorithms with explainable AI (XA) creates a reliable early-warning system capable of supporting targeted academic interventions and improving institutional decision-making.