Investigating the availability of wheelchair-accessible exits of urban rail transit stations: an explainable machine learning approach

Despite improvements in accessibility since the last century, wheelchair users continue to encounter challenges when using urban rail transit (URT), particularly due to the limited availability of wheelchair-accessible exits. Focusing on URT exits, this study aims to identify the factors influencing the provision of wheelchair-accessible exits and examine the resulting impacts on accessibility for wheelchair users. Five cities in the Greater Bay Area, China, including 749 URT stations and 3,360 exits, are included. The random forest model and the SHapley Additive exPlanations (SHAP) method are employed to investigate the contribution of nine variables from both city and station levels. At the city level, the results indicate that both GDP per capita and the number of wheelchair users positively contributed to the availability of wheelchair-accessible exits. At the station level, population density surrounding the station demonstrated an inverse exponential correlation with SHAP values, indicating that stations in densely populated areas are less likely to provide wheelchair-accessible exits. Consequently, wheelchair users need to take considerable detours, averaging 74 %, to reach points of interest within the buffer areas of stations, with the greatest detour observed when accessing financial institutions. The policy and planning implications for achieving universal design are discussed.