Research on the Motivation of Rural Homestead Exit Decision under the Framework of Sustainable Livelihoods: New Discoveries Based on Interpretable Machine Learning

The improvement of the rural homestead system is fundamental to increasing farmers’ property income and achieving rural revitalization in China. Based on the sustainable livelihood framework, this study constructs an analytical framework and employs data from a 2022 field survey of 1,764 rural households across seven provinces. Using interpretable machine learning methods, we evaluate the predictive power of various types of capital characteristics on farmers’ decisions to voluntarily exit their rural homesteads. The results indicate that (1) the Extreme Gradient Boosting (XGBoost) algorithm yields the highest prediction accuracy for homestead exit decisions and significantly outperforms traditional statistical models such as logistic regression; (2) the importance ranking of different capital categories is as follows: human capital > social capital > natural capital > physical capital > psychological capital > financial capital; (3) among the variables, frequency of participation in village collective activities, household non-agricultural income, contracted arable land area, and education level positively influence the likelihood of homestead exit, whereas homestead size and housing area exert a negative influence; (4) the key factors affecting homestead exit vary across pilot and non-pilot villages and between suburban and remote areas. Local governments should prioritize the enhancement of human capital, strengthen social capital networks, optimize the disposal of natural and physical capital, and provide psychological support and institutional guarantees. A differentiated policy approach should be adopted to improve farmers’ willingness to voluntarily exit homesteads, thereby accelerating reform of the rural homestead system.