Spatial Analysis of Bikeshare Ridership With Smart Card and POI Data Using Geographically Weighted Regression Method

Abstract

The primary objective of this paper is to explore the spatial analysis of bikeshare ridership with a consideration of the diversity across different station categories using smart card data and points of interests (POIs) data. The bikeshare trip records were obtained from the Citi Bike system of New York City. The POI data in the vicinity of each station were collected through the Google Places API. K-means clustering method was employed to classify the bikeshare stations into five categories. Then, the geographically weighted regression (GWR) method was applied to establish the relationship between bikeshare ridership and various kinds of influencing factors. To account for the diversity across different station categories, five separate GWR models for each station category were developed and compared with the joint model of all station categories. The results of likelihood ratio test confirmed the superiority and importance of building separate models for each bikeshare station category instead of a joint model. In addition, all the developed bikeshare ridership models were applied to predict the ridership of the newly opened stations in the next year. The results were indicated that the prediction performance of separate bikeshare ridership models was generally better than that of the joint model. The findings of this paper could help transportation agency to develop specific planning and management strategies for each station category of the entire bikesharing system.