A machine learning analysis of density structure and public health risk from the perspective of urban agglomerations a case study of Seoul and Busan

As urbanization and regional integration accelerate, high-density urban agglomerations have become a key form of global development. However, the associated health risks have raised increasing concerns. This study focuses on the urban agglomerations of Seoul and Busan, South Korea, and applies interpretable machine learning to explore the complex relationships between multidimensional density, defined by population, land use, and activity, and health risks, measured by infectious and chronic diseases. First, multiple machine learning models are constructed, and the best performing model is selected based on evaluation metrics. Then, SHAP scores are used to interpret the model at both the global and individual levels, analyzing the importance of key variables, non-linear trends, and their interaction effects. The results show that: (1) interpretable machine learning methods are applicable to health risk research in urban agglomeration contexts, with good transparency and explanatory power; (2) in the Seoul urban agglomeration, medium household density, high household density, student density, and medium residential density are the main characteristics influencing health risks, while in the Busan urban agglomeration, high household density and medium residential density play a dominant role. Overall, the Seoul urban agglomeration has a more diverse range of density types associated with health risks and a more complex variable structure compared to Busan.