Exploring the built environment determinants of traffic congestion with explainable machine learning methods

Traffic congestion has become an urgent problem that needs to be addressed with the rapid urbanization occurring globally. Effective planning of land use configurations is recognized as a structural solution to mitigate urban congestion. However, existing studies may inadequately address how built environment variables directly impact urban congestion, which mostly focus on simplified travel-related indicators, city-level comparisons, and linear relationships. To bridge the gaps, this paper proposes an interpretable machine learning approach to explore the nonlinear associations between urban built environments and traffic congestion, utilizing data collected from various sources. The key findings can be outlined as follows. First, the amenity density (which represents the density of consumption-oriented destinations) emerges as the dominant built environment factor influencing traffic congestion, surpassing conventional variables such as road density and population density. Then, non-linear threshold effects critically shape congestion outcomes, as key factors, including land use diversity, road network density, and metro coverage ratio, only exhibit congestion-mitigating effects beyond a specific level. Finally, the synergistic interplay of public transit network integration, job-housing balance, and high mixed land-use within connected urban spaces is crucial for regional congestion mitigation. By incorporating these insights, traffic planners can develop more effective strategies to alleviate spatially and temporally related traffic congestion within road networks.