Investigating Air Pollution Dynamics in Ho Chi Minh City: A Spatiotemporal Study Leveraging XAI-SHAP Clustering Methodology

Air pollution poses an urgent challenge to public health and ecosystems, particularly in rapidly urbanizing regions. Despite the severity of this issue, there is a lack of robust analytical frameworks capable of identifying key variables and their spatial effects across landscapes. Our study directly addresses this void by applying an innovative supervised clustering approach to air quality data in Ho Chi Minh City (HCMC), Vietnam—a rapidly growing urban area grappling with escalating pollution levels. The analytical model employs Shapley Additive exPlanations (SHAP) to interpret feature importance within tree-based machine learning models, supplemented by the Unified Manifold Approximation and Projection (UMAP) technique to explore intersections between affected areas. We utilize a feature set from Rakholia et al. (2023) as input variables for each target time series, with a focus on answering key questions: What pollutants exert the most influence at different times of day? Which areas of the city are most affected? And can this method effectively pinpoint intersections of pollutant effects? Our results reveal morning traffic congestion predominantly elevates levels of Nitrogen Dioxide (NO2), Humidity, and Carbon Monoxide (CO), while afternoon emissions are significantly impacted by Sulfur Dioxide (SO2), CO, and Ozone (O3) due to solar radiation and industrial activities. Through this research, we expect to contribute to the ongoing discourse on urban air pollution management, highlighting the potential of artificial intelligence-driven tools in environmental research and policy-making.