Interpretable machine learning reveals key determinants of creative thinking among East Asian students: Evidence from PISA 2022

Abstract

This study investigates the key factors and underlying mechanisms influencing creative thinking among East Asian secondary school students using PISA 2022 data. Using a rigorous machine learning approach, we compared five predictive models and identified LightGBM as the most accurate in forecasting creative thinking. We subsequently employed SHAP (SHapley Additive exPlanations), an interpretable machine learning technique, to quantify feature contributions and decompose individual predictions into additive effects, enabling both global and local interpretability. The results revealed three key findings: 1) Academic performance (reading, science, and math performance), ESCS (family economic, social and cultural status), and gender were the five most important predictors; 2) While reading and science performance exhibited a nearly linear positive relationship with creative thinking, ESCS, math performance, empathy and self-efficacy in digital competencies exhibit threshold effects—their positive influence was strongest within optimal value ranges. By contrast, creative family climate and expected educational attainment showed negative relationship with creative thinking; 3) SHAP-based interaction analyses identified the strongest interactions between reading performance and ESCS, science performance and gender, and science performance and ESCS. These findings enhance the theoretical understanding of the multidimensional determinants of creative thinking in East Asian educational contexts and provide researchers and practitioners with empirically grounded insights for designing targeted interventions.