Predicting learning achievement using ensemble learning with result explanation.

Area of Science:

• Educational Technology
• Machine Learning in Education
• Data Science

Background:

• Predicting student learning achievement is vital for reducing dropout rates.
• Existing machine learning models for educational prediction often suffer from bias and lack interpretability.
• This limits their practical application in educational settings.

Purpose of the Study:

• To develop a robust and interpretable predictive framework for learning achievement.
• To combine diverse machine learning algorithms for improved prediction accuracy and reliability.
• To leverage interpretability techniques for actionable insights into student performance.

Main Methods:

• An ensemble learning framework was designed, utilizing six base machine learning models.
• Logistic regression was employed as a meta-learner to construct the final ensemble model.
• SHapley Additive exPlanation (SHAP) was used for model interpretability and feature importance analysis.

Main Results:

• The proposed ensemble model demonstrated superior prediction accuracy compared to traditional machine learning and deep learning models on the XuetangX dataset.
• The ensemble approach significantly outperformed baseline methods in predicting learning achievement.
• SHAP analysis provided clear feature importance, enhancing the model's interpretability and trustworthiness.

Conclusions:

• Ensemble learning offers a powerful approach to accurately predict learning achievement while mitigating bias.
• Model interpretability, achieved through methods like SHAP, is crucial for practical educational applications.
• The findings enable more personalized student support and interventions, potentially reducing dropout rates.