Identifying Street Hotspots Using a Network Kernel Density Estimation

Abstract

Throughout history, police have dedicated their attention to locations with a high incidence of crime and/or disorder. These efforts have become more systematic since the introduction of hotspot policing in the 1990s. A large body of research conducted over the last three decades has shown that focusing police resources on areas characterised by high concentration of crime (hotspots) may have positive preventive effects. If crime hotspots inaccurately reflect the spatial patterning of crime, (intense) criminal activity may not attract police attention. The progress of the crime mapping discipline, spurred by the development of information technology (e.g. GIS) and spatial statistics, enabled crime analysts to identify hotspots more accurately than before. When identifying hotspots, crime analysts use spatial statistical methods which are based on the assumption of a 2D homogeneous Euclidean space. However, in an urban setting, the street network may significantly affect the patterning of crime. Crime may be concentrated along the streets, thus forming linear hotspots. When such linear distribution is analysed, the assumption of homogeneity of the 2D space, which is inherent in “traditional” spatial statistical methods, could lead to a false conclusion. In order to avoid that from happening, these methods should be extended to a network space. This paper explains the specifics of spatial analysis in a network environment and its relevance to the crime analysis practice. Finally, the network extension of one of the most popular methods of hotspot identification, i.e. the kernel density estimation (KDE) method, was applied in the present paper. Having compared the network and planar KDE outputs, the authors concluded that the network-based KDE allowed a more accurate identification of linear hotspots, resulting in a more effective deployment of resources.