Machine learning-based academic performance prediction with explainability for enhanced decision-making in...

Area of Science:

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

Background:

• Higher education institutions increasingly use AI to improve teaching.
• Predicting academic performance is vital for university rankings and student opportunities.
• Challenges exist in performance analysis, quality education, and student evaluation.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting academic performance.
• To compare the generalizability of various regression models on diverse datasets.
• To identify key drivers of student academic success.

Main Methods:

• Ten regression models were employed, including standalone ML and an ensemble voting regression (VR) model.
• Two datasets with different feature sets and sizes were utilized for model evaluation.
• Local interpretable model-agnostic explanations (LIME) and SHapley Additive exPlanations (SHAP) were used for model interpretability.

Main Results:

• Linear Regression performed best among standalone ML models.
• The proposed ensemble VR model achieved superior performance on both datasets.
• VR model achieved an R² of 0.9890 on the first dataset and 0.7716 on the second.
• The VR model demonstrated robustness and adaptability across different academic contexts.

Conclusions:

• Machine learning, particularly ensemble VR, offers a powerful approach to predicting academic performance.
• The study provides actionable insights for educators and policymakers to enhance educational strategies.
• Data-informed decision-making can improve student support and institutional effectiveness.