Spatiotemporal analysis of bike-share demand using DTW-based clustering and predictive analytics

This paper investigates bike-share activities and explores their relationships with neighborhood features, advancing our current knowledge for integrating cycle facilities into urban space to support first/last-mile mobility. To identify distinct demand patterns, bike stations are clustered based on time-series ridership. Measuring the similarity between time-series data in the transportation field should take into account the influence of phase difference because similar demands happening in the morning and in the afternoon should be considered dissimilar. This study uses a weighted Dynamic Time Warping to address this issue by assigning a larger weight to data points with a larger time difference, enabling a more realistic measure to compare ridership. Using bike-share trip data from Citi Bike, we identified eight station clusters, each of which exhibits unique temporal activities. We further integrated the locations of various points of interest to explore the profiles of the clusters. Bike stations that are closer to public transport, commercial buildings and food service establishments are generally more popular, suggesting that incorporating spatial contexts can develop a richer understanding of bike-share usage. In addition, this study goes beyond descriptive analytics by investigating the role of neighborhood features in predicting cluster memberships of bike stations. Our results show that non-tree-based models, such as Support Vector Machines and K-Nearest Neighbors, outperform tree-based models. This study provides valuable insights for both urban planners and bike-share operators. When assessing the potential of new bike-share infrastructure, urban planners can deploy our models to identify the cluster-specific demand pattern based on neighborhood features. Bike-share operators can also utilize our findings to identify neighborhoods that require strategic supply of bikes and parking space that may vary within a day.