Graph Attention Networks Unveil Determinants of Intra- and Inter-city Health Disparity

• DOI: 10.48550/arxiv.2210.10142
• Publication Date: 2022-01-01
• Authors: Chenyue Liu; Chao Fan; Ali Mostafavi
• Keywords: Demographics

Understanding the determinants underlying variations in urban health status is important for informing design and planning, as well public policies. Multiple heterogeneous features could modulate prevalence of diseases across different neighborhoods cities cities. This study examines related to socio-demographics, population activity, mobility, built environment their non-linear interactions examine intra- inter-city disparity four disease types: obesity, diabetes, cancer, heart disease. Features facility density are obtained from large-scale anonymized mobility data. These used training testing graph attention network (GAT) models capture feature spatial interdependence among neighborhoods. We tested five U.S. types. The results show that GAT model can predict people based on top determinant features. findings unveil activity built-environment along with socio-demographic differentiate such a great extent using these high accuracy. also trained one city another accuracy, allowing us quantify similarity discrepancy status. provide novel approaches insights designers, planners, officials better understand improve disparities by considering significant interactions.