Leveraging Machine Learning to Uncover Integrated Impacts of Urban Environment Features on Human Health

Abstract

While human health is significant to sustainable development as cities keep increasing and are estimated to contain 70% of the population by 2050, the quantitative relationship of how the urban environments influence human health remains so poorly understudied due to the convoluted connection and co-effect among numerous urban factors, as well as the large-scale data integration difficulty. Previous researchers studying urban health mainly focus on finding specific urban environment feature’s influence or try to analyze the mechanisms of features co-effect on a coarse scale larger than census tract level or just with qualitative approach. Thus, this research set and reached the objectives to quantify the essential, specific and integrated relationship between urban environment and human health, and specify urban indicators’ co-effect at census tract level mainly from the urban environment’s perspective. With the development of data science, various statistical methods including Machine Learning (ML), SHAP analysis, and Pearson correlation analysis offer promising opportunities for analyzing such complex relationships, facilitated by the increasing availability of urban data by modern tools and the advancements in computational efficiency. By utilizing these methods, this research analyzes 6044 census tracts in 10 US metropolitans (New York City, Los Angeles, Chicago, Houston, Phoenix, Jacksonville, San Francisco Area, Seattle Area, Washington DC Area, Boston Area) with data from 2015 to 2024, builds a framework for census tract level urban health index developed from literature review and WHO urban health indicators framework (2014), quantifies 27 urban environment features’ influence on human health indicators especially on life expectancy, and builds a high performance urban health prediction model for future intervention suggestion for policy makers, making significant meaning towards the urban planning practice.