Revealing key factors of heat-related illnesses using geospatial explainable AI model: A case study in Texas, USA

Abstract

The increasing frequency of extreme weather has led to a notable rise in heat-related health issues. Machine learning algorithms have shown promise in modeling and predicting such outcomes. However, previous studies often neglect spatial components, overlooking the importance of spatial heterogeneity in assessing regional differences in environmental impacts. This study addresses these gaps by employing the geospatial explainable AI (GeoXAI) framework to enhance the spatial interpretability of complex models. The main objective of this study is to understand how geographic location influences factors associated with heat-related emergency department visits (EDVs) across urban and rural areas in Texas. We first leverage automated machine learning (AutoML) to optimize model selection. Then, we employ the GeoShapley approach to analyze the spatial variability of factors contributing to heat-related EDVs. Key findings revealed significant spatial variability and distinct feature importance across urban and rural areas. Socioeconomic and demographic factors were more strongly associated with vulnerability to heat-related health incidents compared to environmental and meteorological variables. Additionally, infrastructure elements, such as transportation systems, were associated with an increased risk of heat in urban areas. These findings highlight the necessity of incorporating geospatial analysis into heat vulnerability assessments to inform targeted public health interventions. By recognizing spatial variability in risk factors, policymakers can implement location-specific strategies to reduce heat-related health burdens, particularly in vulnerable urban communities.