Incorporating multimodality and access modal shares in transit station typology based on the node-place model and machine learning: A case study of Chukyo metropolitan area in Japan

Abstract

This study aims to develop a typology framework for evaluating stations by incorporating multimodal accessibility, multidimensional features, and interpretability using first-mile access patterns. We conducted an exploratory factor analysis on 45 indicators for the 394 stations in the Chukyo Metropolitan Area (CMA) and identified six factors: local place, accessibility, multimodal place, rail node, morphology, and demographics. Six types were subsequently derived using hierarchical clustering, and the stations were evaluated based on the node-place model. We then interpreted and compared station-level feature importance for access modal shares across clusters using Shapley values for Gradient Boosting Decision Tree (GBDT) models. The first main finding was that station areas are supported by expanded catchments beyond walking, with bicycle catchments significantly influencing access modes. Second, a multidimensional and specific approach suggested the potential to advance the previous node-place model framework. Third, the active modes and private vehicles shared a macro-level alignment, but revealed conflicting tendencies with respect to access modes. Fourth, active transit accessibility and number of residents within bicycle catchments showed different nonlinear patterns in relation to access modes, which further varied across clusters. Finally, intersection density tended to discourage private-vehicle access at the macro level, although this was not consistent with the cluster hierarchy. This study is expected to advance station typology research and serve as a useful tool for integrated land use and transport planning.