Identifying and characterizing popular non-work destinations by clustering cellphone and point-of-interest data

Abstract

This paper uses recently newly available datasets to construct a typology of commercial patches reflecting the attraction that specific places hold for individuals and groups and corresponding to the non-work activities that they choose to engage in.For this study, a space-time pattern-detection algorithm was applied to six months of cellphone traces which identified 93 (precise but fluctuating) locations in Singapore with the highest density of people for each hour of the week. Next, we used Google Maps Places Application Programming Interface (API) to web-scrape (or derive information about) establishments for each of the 93 spots identified. A DBSCAN algorithm enabled us to form geometrical clusters for these establishments and produced a new geometry of patches composed of individual establishments. A selection of indicators captured features of the patches' spatial structure: their compactness, density, diversity of activity, the presence of anchor stores, and spatial dependence on proximity to shopping malls. Then, a k-medoids algorithm was used to combine these indicators and form homogeneous groups of commercial patches, thereby identifying for example, strips of restaurants or plazas with anchor supermarkets. The most popular places attracting the greatest number of non-work visitors were dense patches composed of diverse types of businesses.