Abstract:Based on the 2020 CFPS data, this study systematically analyzes the impact of individual, household, and regional characteristics on farmers' non-agricultural entrepreneurship using various machine learning models, including logistic regression, decision trees, random forests, gradient boosting decision trees (GBDT), support vector machines (SVM), and neural networks. The results show that ensemble learning methods outperform traditional models, with random forests and GBDT achieving the best performance. Among all features, household insurance expenditure, household savings, land leasing, internet access, educational attainment, and age emerge as key factors influencing non-agricultural entrepreneurship. Furthermore, partial dependence plots reveal the specific predictive patterns of key features and highlight the significant nonlinearity in the effects of household insurance expenditure, household savings, educational attainment and age. The heterogeneity analysis between urban and rural areas indicates that household insurance expenditure and savings significantly influence non-agricultural entrepreneurship across regions. Additionally, land transfer plays a more prominent role in rural areas, while individual age exerts a stronger influence in urban areas.