Mining network hotspots with holes: A summary of results

Abstract

Given a spatial network and a collection of activities (e.g. crime locations), the problem of Mining Network Hotspots with Holes (MNHH) finds network hotspots with doughnut shaped spatial footprint, where the concentration of activities is unusually high (e.g. statistically significant). MNHH is important for societal applications such as criminology, where it may focus the efforts of officials to identify a crime source. MNHH is challenging because of the large number of candidates and the high computational cost of statistical significance test. Previous work focused either on geometry based hotspots (e.g. circular, ring-shaped) on Euclidean space or connected subgraphs (e.g. shortest path), limiting the ability to detect statistically significant hotspots with holes on a spatial network. This paper proposes a novel Network Hotspot with Hole Generator (NHHG) algorithm to detect network hotspots with holes. The proposed algorithm features refinements that improve the performance of a naïve approach. Case studies on real crime datasets confirm the superiority of NHHG over previous approaches. Experimental results on real data show that the proposed approach yields substantial computational savings without reducing result quality.