Pulse of the World’s Bicycles Towards Data-Driven Methods for the Design and Administration of Bicycle Sharing Systems

Bike sharing systems are emerging all over the world as a cheap, green, and healthy mode of public transit. While there is great variety in their implementation, at their core they are all similar: they consist of bike stations, each containing a number of docking points, from which bicycles may be borrowed and to which they may be returned. These systems suffer from a number of open problems: designing effective bicycle redistribution schemes, optimising the placement of stations, and choosing station capacities, among others. In this project, we use data from 10 such systems to develop and evaluate reusable techniques for their design and administration. Our study is the first to develop and apply uniform analytical methods across several disparate systems, enabling inferences at a previously unavailable level of generality. In particular, we present methods for the unsupervised discovery of naturallyoccurring groups of stations by three separate notions of behaviour. These methods facilitate the design of redistribution vehicle schedules, identification of candidate zones for station addition/removal, and selection of station capacities. We also investigate the prediction of station utilisation at fixed intervals in the future. We demonstrate the utility of our techniques, which require only very simple, lowdimensional, and publicly-available data, by drawing general inferences about the behaviours of bike sharing systems. We discover evidence that relates the size of a system to its behavioural heterogeneity, and evidence to suggest that there is a significant amount of transferable knowledge between systems of similar sizes.