Group anomaly detection for spatio-temporal collective behaviour scenarios in smart cities

Abstract

Group anomaly detection in terms of detecting and predicting abnormal behaviour from entities as a group rather than as an individual, addresses a variety of challenges in spatio-temporal environments like e.g. traffic and transportation systems, smart cities, geoinformation systems, etc. They provide information about a commonly large number of individual entities. Examples for such entities would be airplanes and drones, vehicles, ships but also people, remote sensors and any other information source in interaction with the environment. However, as point anomaly detection is quite common for revealing the abnormal behaviour of individual entities, the collective behaviour of the individuals as a group remains completely uncovered. For example potential for traffic flow optimizations or increased local traffic guideline violations cannot be detected by one single drive but by considering the behavior of a group of vehicle drives in this area. With this work-in-progress we elaborate the potential of group anomaly detection algorithms for spatio-temporal collective behaviour scenarios in smart cities. We describe the group anomaly detection problem in the context of urban planning and demonstrate its effectiveness on a public real-world data set for urban rental bike rides and stations in and around Munich revealing abnormal groups of rides, which allows to optimize the rental bike accessibility to the population and with that to contribute to a sustainable environment.