Relationship between built environment and metro ridership: machine learning analysis

This study examines the driving factors and nonlinear effects of built environment (BE) on passenger flow within the influence areas of rail transit stations. A method is proposed to classify stations and characterize their passenger flow patterns, considering spatial connectivity and transit-oriented development-BE development levels. The Extreme Gradient Boosting (XGBoost) model coupled with the Shapley Additive Explanations value algorithm was employed to identify the key determinants of passenger flow and to explore their nonlinear relationships. Results demonstrate that the XGBoost model outperformed the Multiscale Geographically Weighted Regression (MGWR) model, exhibiting minimal prediction error fluctuations and achieving the highest R 2 value of 0.85. Indicators related to station fineness and service capacity were found to exert the most significant influence on passenger flow. Furthermore, the nonlinear relationship between the BE factors and passenger flow reveals specific thresholds, offering guidance for station spatial layout and the optimization of operational strategies according to station typologies.