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.

Keywords

Suburban railway / Heterogeneous route choice behavior / Latent dirichlet allocation / Interpretable machine learning

References

[1] Tang L, D’Ariano A, Xu X, Li Y, Ding X, Samà M. Scheduling local and express trains in suburban rail transit lines: mixed–integer nonlinear programming and adaptive genetic algorithm. Comput Oper Res, 2021, 135 105436

[2] Bai C, Liu Y, Zhang W, Zhu W, Teng J. Suggestions for optimizing urban rail transport organization based on passenger demand under conditions of four-network integration. Urban Mass Transit (in Chinese), 2024, 27: 1-4

[3] Xu G, Tang Y, Yin Q, Gao F, Zhang R. Thoughts on the early intervention of Shanghai's new suburban railway operation. Urban Mass Transit (in Chinese), 2025, 28: 206-210+217

[4] Chen P, Zhang X, Gao D. Preference heterogeneity analysis on train choice behaviour of high-speed railway passengers: a case study in China. Transp Res Part A Policy Pract, 2024, 188 104198

[5] Peng W, Teng J, Wang H. Understanding heterogeneous passenger route choice in municipal rail transit with express and local trains: an empirical study in Shanghai. Urban Rail Transit, 2024, 10: 122-143

[6] Zhang Y, Yao E, Wei H, Zuo T, Liu S. Constrained multinomial probit route choice modeling for passengers in large-scaled metro networks in China. Transp Res Procedia, 2017, 25: 2389-2399

[7] Zhu Y, Jin J, Wang H. Path-choice-constrained bus bridging design under urban rail transit disruptions. Transp Res Part E Logist Transp Rev, 2024, 188 103637

[8] Lieu SJ, Akar G. Understanding rail users' mode choice behavior for first and last mile travel. J Transp Geogr, 2025, 125 104214

[9] Zhuo S, Zhu X, Shang P, Liu Z, Yao Y, Liao F. Behavior-adaptive sync-flow framework: integrating frequency setting and passenger routing in oversaturated urban rail transit networks. Transp Res Part E Logist Transp Rev, 2024, 189 103659

[10] Zhang R, Yang M, Zhang M, Li H, Peng R. Train service replanning in urban rail transit system: an integrated operation mode that combines express/local and short-turning strategies. Comput Ind Eng, 2025, 203 111018

[11] Le J, Teng J. Understanding influencing factors of travel mode choice in urban-suburban travel: a case study in Shanghai. Urban Rail Transit, 2023, 9: 127-146

[12] Zhu Y, Koutsopoulos HN, Wilson NHM. Inferring left behind passengers in congested metro systems from automated data. Trans Res Part C: Emerg Technol, 2018

[13] Xu X, Xie L, Li H, Qin L. Learning the route choice behavior of subway passengers from AFC data. Expert Syst Appl, 2023, 95: 615-639

[14] Li H, Li X, Xu X, Liu J, Ran B. Modeling departure time choice of metro passengers with a smart corrected mixed logit model - a case study in Beijing. Transp Policy, 2018, 69: 106-121

[15] Teng J, Liu W. Development of a behavior-based passenger flow assignment model for urban rail transit in section interruption circumstance. Urban Rail Transit, 2015, 1: 34-46

[16] Guo J, Tian Y, Li J, He J, Luo Q. Research on passenger route choice behavior when subway single-line operation is interrupted due to emergencies. Urban Mass Transit (in Chinese), 2025, 28: 40-44

[17] Anderson MK, Nielsen OA, Prato CG. Multimodal route choice models of public transport passengers in the Greater Copenhagen Area. EURO J Transp Logist, 2017, 6: 221-245

[18] McFadden D, Train K. Mixed MNL models for discrete response. J Appl Econom, 2000, 15: 447-470

[19] Cheng Y, Ye X, Fujiyama T. How does interchange affect passengers’ route choices in urban rail transit? – a case study of the Shanghai Metro. Transp Lett, 2022, 14: 416-426

[20] Yu C, Li H, Xu X, Liu J. Data-driven approach for solving the route choice problem with traveling backward behavior in congested metro systems. Trans Res Part E: Logist Trans Rev, 2020

[21] Suryakant B, Saeid S, Shervin A. Impact of station design and passengers flow on urban rail dwell time: a systemwide analysis using APC and AVL data. Transp Res Rec, 2024, 2678: 266-280

[22] Mu W, Wang X, Li C, Xiong S. Dynamic modeling for metro passenger flows on congested transfer routes. Mathematics, 2023

[23] Shi Z, Pan W, He M, Liu Y. Understanding passenger route choice behavior under the influence of detailed route information based on smart card data. Transportation, 2025, 52: 615-639

[24] Ali M. Discrete choice models and artificial intelligence techniques for predicting the determinants of transport mode choice—a systematic review. Comput Mater Contin, 2024, 81: 2161-2194

[25] Zhao X, Yan X, Yu A, Hentenryck PV. Prediction and behavioral analysis of travel mode choice: a comparison of machine learning and logit models. Travel Behav Soc, 2020, 20: 22-35

[26] Wagner F, Milojevic-Dupont N, Franken L, Zekar A, Thies B, Koch N, Creutzig F. Using explainable machine learning to understand how urban form shapes sustainable mobility. Transp Res Part D: Transp Environ, 2022, 111 103442

[27] Yin G, Huang Z, Fu C, Ren S, Bao Y, Ma X. Examining active travel behavior through explainable machine learning: insights from Beijing, China. Transp Res Part D: Transp Environ, 2024, 127 104038

[28] Abulibdeh A. Analysis of mode choice affects from the introduction of Doha Metro using machine learning and statistical analysis. Trans Res Interdiscip Perspect, 2023, 20 100852

[29] Zhuang C, Li S, Tan Z, Gao F, Wu Z. Nonlinear and threshold effects of traffic condition and built environment on dockless bike sharing at street level. J Transp Geogr, 2022, 102 7690

[30] Sun Y, Dong Y, Waygood EOD, Hamed N, Jiang Y, Chen Y. Machine-learning approaches to identify travel modes using smartphone-assisted survey and map application programming interface. Trans Res Record: J Trans Res Board, 2022

[31] Mahdi R, Shaun SW, Amirarsalan MM, Khaled K. Application of Bayesian ordinal logistic model for identification of factors to traffic barrier crashes: considering roadway classification. Transp Lett, 2021, 13: 308-314

[32] Hamed N, E O D Waygood, Zachary P, Wang B(2024) Which variables influence electric vehicle adoption? Transportation: https://doi.org/10.1007/s11116-024-10525-1.

[33] Hamed N, Ciari F, Isler CA. Influence of pedestrianization on travel behavior changes: A case study of Montreal. J Trans & Health, 2025

[34] Hatami F, Rahman MdM, Nikparvar B, Thill JC. Non-linear associations between the urban built environment and commuting modal split: a random forest approach and SHAP evaluation. IEEE Access, 2023, 11: 12649-12662

[35] Wardman M, Bonsall PW, Shires JD. Driver response to variable message signs: a stated preference investigation. Trans Res Part C: Emerg Technol, 1997, 5: 389-405

[36] Bogers EAI, Viti F, Hoogendoorn SP. Joint modeling of advanced travel information service, habit, and learning impacts on route choice by laboratory simulator experiments. Transp Res Rec, 2005, 1926: 189-197

[37] Shin H, Kim D, Kho S, Cho S. Valuation of metro crowding considering heterogeneity of route choice behaviors. Transp Res Rec, 2020, 2675: 162-173

[38] Wang Y, Wang Y, Choudhury C. Modelling heterogeneity in behavioral response to peak-avoidance policy utilizing naturalistic data of Beijing subway travelers. Trans Res F: Traffic Psychol Behav, 2020, 73: 92-106

[39] Chen Z, Zhang Y, Wu C, Ran B. Understanding individualization driving states via latent Dirichlet allocation model. IEEE Intell Transp Syst Mag, 2019, 73: 41-53

[40] Lai J, Wang Y, Yang Y, Wu X, Zhang Y. Exploring the built environment impacts on online car-hailing waiting time: an empirical study in Beijing. Comput Environ Urban Syst, 2025, 115 102205

[41] Tang M, Cao J, Gong D, Xue G. Unsupervised learning-based exploration of urban rail transit passenger flow and travel pattern mining. Int J Comp Commun & Control, 2024

[42] Liu H, Shi H, Yuan T, Fu S, Ran B. Bus travel feature inference with small samples based on multi-clustering topic model over internet of things. Future Gen Compute Syst, 2025

[43] Çallı L, Çallı F. Understanding airline passengers during COVID-19 outbreak to improve service quality: topic modeling approach to complaints with Latent Dirichlet Allocation algorithm. Transp Res Rec, 2022, 2677: 656-673

[44] Aminpour N, Saidi S. Unveiling mobility patterns beyond home/work activities: a topic modeling approach using transit smart card and land-use data. Travel Behav Soc, 2025, 38 100905

[45] Chen T, Carlos G. XGBoost: a scalable tree boosting system. Knowl Discov Data Mining, 2016

[46] Benalaya N, Amdouni I, Adjih C, Laouiti A, Saidane LA. Deep reinforcement learning approach for UAV search path planning in discrete time and space. Int Wireless Commun Mob Comput (IWCMC), 2024

[47] Zhang X, Zhou Z, Xu Y, Zhao X. Analyzing spatial heterogeneity of ridesourcing usage determinants using explainable machine learning. J Transp Geogr, 2024, 114 103782

[48] Lundberg S, Lee S. A unified approach to interpreting model predictions. Neural Inform Proc Syst, 2017, 30: 4768-4777

[49] Friedman JH. Greedy Function Approximation: A Gradient Boosting Machine. Ann Stat, 2000, 29: 1189-1232

[50] Cheng Z, Sun D, Zhao Y, Peng H. Investigating the factors influencing intercity travel mode choice in urban agglomerations: insights from a three-phase framework. Transp Res Part A Policy Pract, 2025, 199 104577

[51] Li L, Gao T, Yu L, Zhang Y. Applying an integrated approach to metro station satisfaction evaluation: a case study in Shanghai, China. Int J Transp Sci Technol, 2022, 11: 780-789

[52] Zhou F, Li C, Huang Z, Xu R, Fan W. Fare incentive strategies for managing peak-hour congestion in urban rail transit networks. Transportmetrica A Transp Sci, 2022, 18: 166-187

[53] Martín-Baos JA, López-Gómez JA, Rodriguez-Benitez L, Hillel T, García-Ródenas R. A prediction and behavioural analysis of machine learning methods for modelling travel mode choice. Transp Res Part C Emerg Technol, 2023, 156 104318

Funding

National Natural Science Foundation of China(52402388)

Fundamental Research Funds for the Central Universities

RIGHTS & PERMISSIONS

The Author(s)