Decoding student cognitive abilities: a comparative study of explainable AI algorithms in educational data mining.

Area of Science:

• Educational Data Mining
• Artificial Intelligence in Education
• Cognitive Science

Background:

• Understanding factors influencing student cognitive abilities is crucial for educational advancement.
• Educational data mining (EDM) offers powerful tools for analyzing learning processes.
• The interpretability of AI models in education remains a significant challenge.

Purpose of the Study:

• To investigate factors affecting students' cognitive abilities using AI models.
• To compare the performance of various explainability algorithms in educational data analysis.
• To explore the impact of self-perception and parental expectations on cognitive development.

Main Methods:

• Employed five data-driven artificial intelligence (AI) models for educational data.
• Utilized four interpretable AI algorithms (feature importance, Morris Sensitivity, SHAP, LIME) for global model interpretation.
• Conducted Propensity Score Matching (PSM) causal tests to identify influential factors.

Main Results:

• Self-perception and parental expectations were consistently identified as significant factors influencing cognitive abilities across all tested algorithms.
• Different explainability algorithms (Morris Sensitivity, SHAP, feature importance, LIME) exhibited distinct feature importance rankings, indicating varying analytical perspectives.
• Discrepancies in highlighted features underscore the nuanced interpretations provided by each explainability method.

Conclusions:

• Interpretable AI algorithms provide valuable insights into educational data mining.
• The choice of explainability algorithm can influence the interpretation of factors affecting student cognitive abilities.
• This study demonstrates the practical utility of explainable AI for refining educational applications and research.