Nonlinear and Spatially Varying Impacts of Natural and Socioeconomic Factors on Multidimensional Human Health: A Geographically Weighted Machine Learning Approach

Ensuring healthy lives is both crucial for residents’ quality of life and a core objective of the Sustainable Development Goals. While numerous studies have examined the drivers of human health, few have distinguished between average health and extreme health outcomes. Moreover, the nonlinear effects of various determinants on multidimensional human health remain underexplored. This study aims to investigate the spatially varying contributions of natural and socioeconomic factors to multidimensional human health, focusing particularly on their nonlinear relationships. We first quantified multidimensional human health at the prefectural level of China, including average health (Life Expectancy), extreme health (Longevity Index and Centenarian Index), and subjective health (Self-Rated Health). We then combined the Geographically Weighted Random Forest model with the SHAP interpretations to investigate the impacts of natural and socioeconomic factors. Through the GWRF-SHAP framework, we estimated the nonlinear relationships between the four health indicators and their key influencing factors. We found that: (1) The four health indicators exhibited significant positive correlations, except for the relationship between Centenarian Index and Self-Rated Health. (2) Extreme health outcomes (Longevity and Centenarian Index) were predominantly influenced by natural factors, whereas average and subjective health (Life Expectancy and Self-Rated Health) were more strongly associated with socioeconomic conditions. (3) The dominant determinants of human health varied across regions, but socioeconomic factors generally showed stronger influences in northwestern China. (4) Both socioeconomic and natural factors exhibited nonlinear effects and threshold behaviors on health outcomes. These findings suggest that improving socioeconomic conditions is beneficial for enhancing both average and subjective health, whereas managing the natural environment is crucial for promoting extreme levels of health. Our study advocates a multidimensional, spatially tailored, and threshold-sensitive approach to health management.