The multi-parameter optimized belief rule base for predicting student performance with interpretability.

Area of Science:

• Educational Data Mining
• Artificial Intelligence in Education

Background:

• Accurate student performance prediction is vital for personalized learning and educational equity.
• Traditional Belief Rule Base (BRB) models offer interpretability but face challenges like rule explosion and reduced transparency post-optimization.
• Expert knowledge limitations in BRB can also impact prediction accuracy.

Purpose of the Study:

• To develop an interpretable student performance prediction model.
• To enhance the accuracy and interpretability of Belief Rule Base (BRB) models.
• To address limitations of existing BRB models in handling numerous attributes and expert knowledge constraints.

Main Methods:

• Implemented an attribute selection method using Random Forest to identify key performance-influencing features.
• Defined specific criteria for maintaining interpretability throughout the model optimization process.
• Developed a multi-parameter optimization technique with interpretable constraints for the belief rule base model.

Main Results:

• The proposed interpretable student performance prediction model (IBRB-m) demonstrated effectiveness in a case study.
• Attribute selection successfully filtered important features, mitigating the rule combination explosion problem.
• The multi-parameter optimization with interpretable constraints improved both prediction accuracy and model transparency.

Conclusions:

• The IBRB-m model offers a viable solution for accurate and interpretable student performance prediction.
• The integration of attribute selection and constrained optimization enhances the practical applicability of BRB in educational settings.
• This approach supports informed teaching decisions and personalized learning pathways.