Detecting (unusual) events in urban areas using bike-sharing data

Abstract

Social media, traffic sensors, GPS trajectories, and location-based social network data provide diverse spatio-Temporal information sources that help to detect and analysis spatio-Temporal events. Nowadays, bike sharing systems are active all over the world in major cities, and collecting a large amount of data regarding trips taken by users and status of the stations. Through analysis of the data aggregated by bike sharing systems, one can gain an understanding of crowd/commuter movements and behaviors. However, no one has used only the bike sharing data for generic event detection. In this paper, we propose a clustering-based detection method to identify spatiotemporal events that deviate from normal or regular everyday life using publicly available bike sharing data. In particular, we apply spectral clustering on bike station and bike flow data as evolving graphs and monitor changes of the bike share network (edge/node values) over time. Our proposed method decides whether a cluster is expected or anomalous (unusual). When a cluster is anomalous, there is an unusual event occurring at that time instance. Preliminary results on 6-months of data from Philadelphia and Washington DC are used to show the feasibility of our proposed method. In particular, our preliminary results show that some signatures of local (and less prominent) events (e.g., university events/activities in an urban area) can show up when bike sharing data is utilized for generic event detection.