Performance Evaluation of the Full Transfer Process between High-Speed Rail and Metro in Railway Hubs: A Simulation and Machine Learning Approach

High-speed railway (HSR) hubs are pivotal nodes in the railway transportation network, and station performance during peak periods is crucial in maintaining transfer efficiency and passenger safety. This study proposes a research framework integrating agent-based simulation, metamodeling, and interpretable machine learning [extreme gradient boosting (XGBoost) with Shapley additive explanation (SHAP)]. The framework explicitly models heterogeneous passenger types and transfer facilities, generating large-scale multiscenario data sets. A surrogate model is then constructed to capture nonlinear relationships among passenger flow, facility characteristics, and transfer performance. This methodological integration represents a novel contribution by combining microscopic simulation with interpretable artificial intelligence (AI) for railway hub transfer analysis. The proposed framework is validated through a case study of Nanjing South Railway Station in China. Key findings indicate that transfer efficiency is mainly influenced by passenger flow volumes and the availability of automatic fare gates (AFGs), with facility quantity exerting a more significant impact than service capability.