Mining Bike-Share Data

This paper studies methods for processing bike-share datasets for the purpose of extracting information that can assist riders, bike-share program designers, city planners, and others. Bike-share datasets describe how shared bicycles are used in an urban environment. They vary considerably in composition and coverage but typically include information such as the locations (bicycle racks) of origin and destination, timestamps, and identifiers for bicycles and riders. This paper provides methods for visualizing such data in a manner that distills useful patterns and for using the data to predict usage. In order to overcome the difficulty in generating meaningful clusters using conventional methods, it presents a novel method of clustering that uses graph condensations. It describes an experimental study of these methods using a publicly available dataset from a popular bike-share program. Results include the encouraging visualizations using graph condensations, significant benefits of some computed features such as trip durations, and some insights for planning related to periodicity and the effects of a pandemic such as COVID-19.