Revealing the impacts of the built environment factors on pedestrian-weighted air pollutant concentration using automated and interpretable machine learning

Urban air pollution poses significant health risks, especially to pedestrians due to their proximity to pollutants and lack of physical protection. Understanding the influence of built environment factors is essential to mitigate this pollution and safeguard pedestrian health. However, most existing literature focus primarily on pollutant sources and dispersion dynamics, paying less attention to the factors that affect the extent of pedestrian exposure to pollutants. Additionally, while machine learning has gained traction in urban studies, challenges remain in model optimization and interpretability, leading to limited transparency and reduced clarity in environment strategy development. To address these gaps, this study proposes a methodological framework to measure pedestrian-weighted air pollutant concentrations (PWAPC) and analyze the complex effects of the built environment. The objectives include (1) integrating air pollution and pedestrian volume data to quantify PWAPC levels, and (2) employing automated machine learning (AutoML) and interpretable machine learning (IML) to model PWAPC and evaluate key built environment impacts. A case study on PM2.5 concentrations in Central London demonstrates the efficiency of AutoML in algorithm selection and hyperparameter optimization. Using IML, critical factors such as points of interest (POIs), traffic infrastructure, diversion ratios, betweenness centrality, street canyon effects, and urban greenness are identified. The analysis also reveals non-linear relationships between these factors and PWAPC. This study provides actionable insights for urban planning and environmental management.