Selective spatial production of urban village regeneration in Guangzhou, China: An index-based XGBoost-SHAP approach

Urban villages are central to stock-based spatial governance in megacities and offer a typical context for examining spatial production theory. This study investigates 285 urban villages in Guangzhou, China. It constructs a 20-indicator Spatial Production Intensity Index (SPII) from multi-source spatial data and combines GIS, spatial autocorrelation analysis, eXtreme Gradient Boosting (XGBoost), and SHapley Additive exPlanations (SHAP) to examine spatial patterns, implementation sequencing, and predictive associations with planning importance. Three findings emerge. First, both the SPII and planning importance exhibit significant positive spatial clustering, with high-value clusters concentrated in the urban core and a clear decline towards the periphery, while planned implementation follows a phased and flexible sequence. Second, planning and development intensity is the strongest predictor of planning importance, demonstrating the guiding role of institutional representations of space. Third, spatial production factors show nonlinear and context-dependent interactions: infrastructure, socio-labour carrying capacity, and cultural-symbolic visibility can amplify, compensate for, constrain, or reshape institutional development potential. These findings reveal urban village regeneration as a process of selective spatial production and provide evidence for differentiated regeneration pathways.