Detecting Isodistance Hotspots on Spatial Networks: A Summary of Results

Spatial hotspot detection aims to find regions of interest with statistically significant high concentration of activities. In recent years, it has presented significant value in many critical application domains such as epidemiology, criminology and transportation engineering. However, existing spatial hotspot detection approaches focus on either on Euclidean space or are unable to find the entire set of hotspots. In this paper, we first formulate the problem of Network Isodistance Hotspot Detection (NIHD) as finding all sub-networks whose nodes and edges are reachable from a activity center and have significantly high concentration of activities. Then, we propose a novel algorithm based on network partitioning and pruning (NPP) which overcomes the computational challenges due to the high costs from candidate enumeration and statistical significance test based on randomization. Theoretical and experimental analysis show that NPP substantially improves the scalability over the baseline approach while keeping the results correct and complete. Moreover, case studies on real crime datasets show that NPP detects hotspots with higher accuracy and is able to reveal the hotspots that are missed by existing approaches.