Using machine learning to predict student mathematics performance in six East Asian countries: evidence from PISA...

Area of Science:

• Educational Psychology
• Artificial Intelligence in Education
• International Comparative Education

Background:

• Mathematics performance is crucial for younger generations in the AI era.
• East Asian students' mathematical achievement is notable, yet determinants are under-researched using machine learning.
• Limited studies explore combined factors influencing math success in East Asian contexts via advanced computational methods.

Purpose of the Study:

• To evaluate six machine learning models for predicting mathematics performance.
• To identify the most accurate algorithm for analyzing PISA 2022 data from East Asian students.
• To pinpoint key predictors of mathematics achievement in high-performing East Asian economies.

Main Methods:

• Six machine learning models (Random Forest, LightGBM, XGBoost, AdaBoost, Elastic Net, Linear Regression) were evaluated.
• A sample of 26,969 fifteen-year-old students from PISA 2022 was analyzed.
• Feature selection identified 24 key predictors; SHapley Additive exPlanations (SHAP) quantified predictor influence and interactions.

Main Results:

• XGBoost was the optimal model, achieving R² = 0.5758 and explaining 57.03% of variance in math achievement.
• Mathematics self-efficacy was the strongest predictor, followed by extracurricular activity participation and weekly math instruction time.
• Affective, behavioral, and instructional factors demonstrated higher predictive importance than structural and socioeconomic variables.

Conclusions:

• Student-proximal determinants significantly influence mathematics achievement in Confucian Heritage Culture educational settings.
• Findings support prioritizing self-efficacy, optimizing instructional time, and fostering equitable learning environments.
• The study offers theoretical contributions and directions for future educational research and policy.