Explainable AI for Predicting and Understanding Mathematics Achievement: A Cross-National Analysis of PISA 2018

Machine learning models revealed non-linear patterns in math achievement: The study found that more complex models like Random Forest and Artificial Neural Networks significantly outperformed traditional linear regression, suggesting that student math performance is shaped by complex interactions between factors rather than simple additive relationships.

Socio-economic status remains a dominant predictor across countries: Among the key factors identified using explainable AI techniques (SHAP values and feature importance analysis), socio-economic status emerged as a consistent predictor of mathematics achievement across the ten countries studied, with 67,329 students analyzed, underscoring persistent equity challenges in education.

Context matters: predictors' importance varies by country: While factors like study time, teacher motivation, and student attitudes toward mathematics were identified as important, their relative impact differed across the ten nations examined, demonstrating that one-size-fits-all educational policies may be ineffective and that reforms should account for local contexts.

Explainable AI made complex models trustworthy for policy use: By using interpretability tools like SHAP values and decision tree visualizations, the researchers demonstrated that high-performing machine learning models don't have to be "black boxes"—policymakers can understand why predictions are made, making AI-driven insights more actionable for designing targeted educational interventions.

Personalized learning strategies are now data-informed: The study's findings support moving beyond broad demographic categories toward understanding individual student factors (attitudes, study habits, teacher quality) that influence math achievement, enabling schools to develop more tailored support strategies.