Impact of Subway Passenger Flow–Built Environment Interaction on Road Congestion around Subway Stations Using Machine-Learning Models

Abstract To alleviate urban traffic congestion, metro systems have become critical urban transportation infrastructure. While effectively relieving pressure on long-distance commutes, the areas surrounding metro stations often emerge as congestion-sensitive zones due to the transient aggregation of feeder traffic and the high-density land use patterns concentrated there. This study focuses on road networks within a 500 m radius of stations along six metro lines in Kunming, China. It integrates four categories of multisource heterogeneous data: metro smart card transactions; point of interest (POI) data; meteorological data; and the Amap Road Congestion Index. Employing an innovative integrated modeling framework combining light gradient boosting machine (LightGBM) and Shapley additive explanations (SHAP) interpretability analysis, this research dissects the complex interactive relationships underlying traffic congestion around metro stations. The research results indicate that the LightGBM model can effectively identify road traffic congestion conditions around metro stations, and SHAP can effectively analyze the nonlinear relationship of the interaction between metro passenger outflow and POI facility density on road traffic congestion around metro stations. The study found that, when the density of pedestrian facilities exceeds 60 / km 2 , as the density of pedestrian facilities increases, its impact on road traffic congestion becomes positive, with an effect value above 0. When the density of government institutions exceeds 120 / km 2 and passenger outflow is greater than 120 people/10 min, the interaction between the two further exacerbates road traffic congestion. When company density is around 50 / km 2 , the SHAP value spikes sharply, reaching a maximum exceeding 0.6, indicating that, in areas with moderate company density but extremely high passenger flow, the interaction between the two generates a substantial positive predictive driving force for traffic congestion. When the density of business residential areas is low, it exerts an inhibitory effect on road traffic congestion; however, as its value increases, this inhibitory effect weakens and gradually transitions to a positive contribution. Further, regarding whether it is a peak period, the influence of off-peak periods on road traffic is compactly clustered around zero; however, once entering a peak period, the feature contribution exhibits a stepwise increase. These findings enhance the understanding of the causes of traffic congestion near urban metro stations, providing a robust empirical basis for optimizing functional zoning around stations, implementing differentiated facility allocation, and enabling precise traffic control strategies.