Abstract: Based on a sample of 33, 569 teachers from 334 kindergartens, primary and secondary schools in Shangqiu, Zhoukou and Xinyang, Henan Province, an interpretable machine learning model is constructed to predict occupational burnout among basic education teachers. By comparing logistic regression, random forest, XGBoost, CatBoost and LightGBM algorithms, CatBoost demonstrated the best comprehensive performance (AUC=0.88, specificity=0.86). Demographic analysis revealed that burnout rates were higher among teachers in villages and towns compared to those in districts/counties, and unmarried teachers and teachers with low economic status faced a significantly higher risk. SHAP explanatory analysis identified and ranked the important risk and protective factors for burnout among basic education teachers. The top five predictors were: deep acting, job involvement, role overload, depression, and neuroticism, with deep play and job involvement serving as the core protective factors, and role overload, depression, and neuroticism acting as the main risk sources. An interpretable prediction model for teacher burnout is constructed, with'deep acting-job involvement'identified as the dual core protective factors and'role overload-depression-neuroticism'as the triple core risk sources, thereby providing evidence for regional education management departments to implement tiered interventions.