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.
|Original language||English (US)|
|Title of host publication||Advances in Spatial and Temporal Databases - 15th International Symposium, SSTD 2017, Proceedings|
|Editors||Wei-Shinn Ku, Agnes Voisard, Haiquan Chen, Chang-Tien Lu, Siva Ravada, Matthias Renz, Yan Huang, Michael Gertz, Liang Tang, Chengyang Zhang, Erik Hoel, Xiaofang Zhou|
|Number of pages||19|
|State||Published - 2017|
|Event||15th International Symposium on Spatial and Temporal Databases, SSTD 2017 - Arlington, United States|
Duration: Aug 21 2017 → Aug 23 2017
|Name||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Other||15th International Symposium on Spatial and Temporal Databases, SSTD 2017|
|Period||8/21/17 → 8/23/17|
Bibliographical noteFunding Information:
Acknowledgement. This material is based upon work supported by the National Science Foundation under Grant No. 1029711, IIS-1320580, 0940818, IIS-1218168, and IIS-1541876, the USDOD under Grant No. HM0210-13-1-0005, the ARPA-E under Grand No. DE-AR0000795, and the University of Minnesota under the OVPR U-Spatial. We are particularly grateful to Kim Koffolt and the members of the University of Minnesota Spatial Computing Research Group for their valuable comments.