Open source data–driven method to identify most influencing spatiotemporal factors. An example of station–based bike sharing

Abstract The development of information and telecommunication technologies has allowed the collection of vast amounts of data. Many organizations slowly open their data, providing heterogeneous data that can potentially improve our understanding of travel behavior. Along the lines of an increasing volume of spatial data for transport applications and the lack of a framework for their utilization, this research presents a data–driven approach which identifies the most influencing factors on the ridership of stations–based transportation systems, with a focus on bike sharing. Through the collect and processing of open source spatial information, we estimate models for bike-sharing ridership. The models are built in different temporal scales and consider stations in multiple cities which form a pooled dataset. The methods are applied for estimating models from five cities in Germany and validated in a sixth city. Arrivals and departures were correlated with 169 spatial variables, which were selected automatically out of 800 variables. The models identified included variables related to leisure activities, residential and commercial land use, university areas, natural environment, cycling infrastructure, touristic points of interest and public transport stations.