A unified clustering approach for identifying functional zones in suburban and urban areas

It is well known that a modern city comprises multiple functional zones such as business, entertainment and shopping areas. It is also well established that identifying functional zones can effectively assist in understanding human mobility behavior and planning urban development. Less is known about suburban functional zones. In this paper we propose a unified approach that successfully identifies functional zones in both suburban and urban areas. Our approach is driven by POIs (points of interest) without relying on human movement trajectories. We offer a two-stage clustering algorithm, for which the key is computing geolocation clusters based on the density and physical location of POIs. Often in the literature geolocation clusters are POI independent, e.g. derived from existing road networks or cell coverage map. In contrast, our geoclusters can be of varying sizes, arbitrary shapes and different POI densities. Furthermore, we observe that a sizable fraction of our geoclusters have a small number of dominating functionalities. This illustrates that suburban functional zones exist and makes it possible to further group geoclusters by their functional characteristics into meaningful functional zones. To validate our approach, we carry out detailed studies of a suburban area in New Jersey and the Manhattan area of New York City. We share our insights from these studies.