Understanding metro ridership drivers is vital for transit planning. This study employs a multidimensional framework–encompassing the built environment, intermodal interaction, network topology, and operational features–to analyze ridership at 100 Guangzhou metro stations, focusing on two peak days: 31 December 2024 (12.2 million riders) and 1 January 2025 (national record). Three models (OLS, XGBoost, LightGBM) are tested, with LightGBM selected. SHAP is used to interpret the model and identify key variables. Results show that operational features, especially the capacity coefficient (CO), exert the greatest influence, surpassing traditional factors. Other significant predictors include commercial area (CA), parking lots (PL), and betweenness centrality (BC). The study also introduces a novel application of Line 11 as a spatial boundary for defining the downtown area. These findings provide actionable insights for enhancing metro operations, capacity, and multimodal integration, demonstrating the utility of machine learning and SHAP-based interpretation in understanding metro ridership under extreme conditions.
Bingjie Zhan (Thu,) studied this question.