Mining Multimodal Travel Patterns of Metro and Bikesharing Using Tensor Decomposition and Clustering

Multimodal transportation systems, particularly those combining metro and bikesharing, have become central to addressing the first- and last-mile connectivity challenges in urban environments. This study presents a comprehensive data-driven framework to analyze the spatiotemporal interplay between metro and dockless bikesharing usage using real-world data from Tianjin, China. Two primary methods are employed: K-means clustering is used to categorize metro stations and bike usage zones based on temporal demand features, and non-negative Tucker decomposition is applied to a three-way tensor (day, hour, station) to extract latent mobility modes. These modes capture recurrent commuting and leisure behaviors, and their alignment across modes is assessed using Jaccard similarity indices. Our findings reveal distinct usage typologies, including mismatched (misalignment of jobs and residences), employment-oriented, and comprehensive zones, and highlight strong temporal coordination between metro and bikesharing during peak hours, contrasted by spatial divergence during off-peak periods. The analysis also uncovers asymmetries in peripheral stations, suggesting differentiated planning needs. This framework offers a scalable and interpretable approach to mining multimodal travel patterns and provides practical implications for station-area design, dynamic bike rebalancing, and integrated mobility governance. The methodology and insights contribute to the broader effort of data-driven smart city planning, especially in rapidly urbanizing contexts.