Unraveling the Spatiotemporal Drivers of Sustainable Human Settlement Quality in China: Evidence from Explainable Machine Learning and Panel Econometrics

Human settlement quality is a key dimension of sustainable urban development, yet its spatiotemporal evolution and associated mechanisms remain insufficiently understood, particularly under rapid urbanization and regional inequality. This study aims to evaluate the Human Settlement Quality Index (HSQI) across 31 Chinese provinces from 2012 to 2021 and to examine its nonlinear predictive patterns, average conditional associations, and region-specific pathways. A composite HSQI was constructed using an entropy-weighted multi-criteria decision-making framework based on 25 indicators covering natural, human, social, residential, and supporting systems. XGBoost-SHAP was used to identify global feature importance and nonlinear predictive patterns, while a two-way fixed effects panel model and regional group regressions were employed to estimate average conditional associations and regional heterogeneity. The results show that China’s HSQI increased by 12.22% from 2012 to 2021, with an initial decline followed by sustained improvement and narrowing regional disparities. Per capita GDP was the most important predictive factor, while human and supporting systems jointly accounted for more than 60% of the total feature importance. Several core indicators exhibited nonlinear threshold-like response patterns, and regional association patterns differed substantially. Eastern China showed signs of a weaker association between economic growth and HSQI improvement, central China showed stronger associations with digital logistics and infrastructure, and western China remained closely linked to ecological foundation protection. These findings demonstrate the complementarity of interpretable machine learning and panel econometrics and provide evidence for differentiated, sustainability-oriented strategies to improve human settlement quality.