Academic achievement prediction in higher education through interpretable modeling.

Area of Science:

• Educational Technology
• Machine Learning in Education
• Data Science in Higher Education

Background:

• Student academic achievement is a key metric for educational quality.
• Predicting achievement aids educators in tailoring instruction and improving student outcomes.
• Challenges exist in extracting actionable insights from educational data using traditional methods.

Purpose of the Study:

• To introduce a novel machine learning approach, the XGB-SHAP model, for predicting student academic achievement.
• To address limitations of traditional algorithms in delineating factors influencing student grades.
• To enhance the accuracy of student performance prediction in higher education.

Main Methods:

• The study employed the Extreme Gradient Boosting (XGBoost) algorithm combined with SHapley Additive exPlanations (SHAP).
• The XGB-SHAP model was applied to a dataset of 87 university students in a Japanese course (Sept 2021 - June 2023).
• Model performance was compared against three other machine learning models.

Main Results:

• The XGB-SHAP model demonstrated high accuracy, achieving a Mean Absolute Error (MAE) of approximately 6 and an R-squared value of 0.82.
• The model outperformed three other machine learning models in prediction accuracy.
• The analysis revealed how different instructional modes impact factors contributing to student achievement.

Conclusions:

• The XGB-SHAP model offers a superior approach to predicting student academic achievement.
• Customized feature selection based on teaching modes is necessary for effective prediction.
• Integrating self-directed learning skills into predictive indicators is vital for accurate performance forecasting.