Understanding Route Choice Behavior in Composite Urban Rail Transit Network with Interpretable Machine-Learning Approach: A Case Study in Shanghai, China

Abstract Analyzing route choice behavior and understanding the heterogeneous impacts on passenger decisions are key to improving transportation service quality in integrated suburban railway and metro networks. While existing studies focus on metro systems using Mixed Logit (MXL) models or clustering methods with Multinomial Logit models, these approaches struggle to capture the diverse factors in travel scenarios and the complex nonlinear relationships in decision-making. This study integrates Latent Dirichlet Allocation with Machine-Learning models like eXtreme Gradient Boosting (XGBoost), Support Vector Machine, and Random Forest to analyze passenger behavior within Shanghai’s composite rail network. It identifies key influencing factors such as socio-economic demographic characteristics, travel purposes, and the express-to-local departure ratio. XGBoost outperforms other models, including the traditional MXL, in predictive performance. The study highlights significant heterogeneity in route choices across different passenger groups, underscoring the need for personalized transportation solutions. Based on these findings, this study proposes the following actionable suggestions for suburban railway in the composite network: In terms of operational optimization, it is proposed to add express train overtaking stations in key commuting corridors and optimize the timetable to reduce transfer waiting times, thereby improving overall travel efficiency. Besides, time-of-day differentiated pricing and combined-ticket discounts are proposed to improve the rationality of the ticket-price structure. Finally, it is recommended to enhance the personalized route recommendation system.