What Really Drives Teachers to Learn? Insights from an Explainable Machine Learning Approach

<title>Abstract</title> <p>Teacher professional learning is a cornerstone of school effectiveness and improvement, yet many teachers lack autonomous motivation for continuous professional development (CPD). Traditional linear models may miss complex, non-linear associations among psychological and workplace conditions. Grounded in Self-Determination Theory (SDT) and the school effectiveness research tradition, this study applies explainable machine learning to survey data from 472 school teachers in China. Among the models compared, XGBoost showed the strongest cross-validated predictive performance (AUC = 0.851). SHAP analyses indicated that school leadership and teachers' learning conceptions were among the most influential predictors in the model. Subgroup analyses further suggested two patterns of heterogeneity: the predictive contribution of teaching experience was lower in the transformative-leadership subgroup than in the conservative-leadership subgroup, and under high work pressure, task autonomy and self-efficacy became more salient while leadership and colleague support showed lower predictive contributions. Because the study is cross-sectional and model-based, these findings should be interpreted as predictive patterns that can inform future causal and longitudinal research and do not provide evidence of causal effects.</p>