Network-Based Space-Time Scan Statistics for Detecting Micro-Scale Hotspots

Abstract

Events recorded in urban areas are often confined by the micro-scale geography of street networks, yet existing spatial–analytical methods do not usually account for the shortest-path distance of street networks. We propose space–time NetScan, a new spatial–temporal analytical method with improved accuracy for detecting patterns of concentrations across space and time. It extends the notion of a scan-statistic-type search window by measuring space-time patterns along street networks in order to detect micro-scale concentrations of events at the street-address level with high accuracy. Performance tests with synthetic data demonstrate that space-time NetScan outperforms existing methods in detecting the location, shape, size and duration of hotspots. An empirical study with drug-related incidents shows how space-time NetScan can improve our understanding of the micro-scale geography of crime. Aside from some abrupt one-off incidents, many hotspots form recurrent hotbeds, implying that drug-related crimes tend to persist in specific problem places.