Most significant hotspot detection using improved particle swarm optimizers

Abstract

Significant circular hotspot detection (SCHD) aims at identifying those circular regions in a spatial space where the occurrence of a particular activity is uncommonly higher than the surrounding areas. Plentiful societal applications make SCHD a problem of utmost interest. In many domains, detection of the most significant circular hotspot (MSCHD) is of further usefulness. Well-timed detection of the most significant hotspot helps crucially for short term response planning (STRP) in situations like a disease outbreak, police vigilance, etc. State of the art methods like SaTScan, identify circular hotspots by listing all possible circles in the search area, followed by a statistical significance test, making it a very high computational cost problem. Considering their high costs, these methods are inefficient in applications related to STRP. To reduce the computational time of SaTScan, two randomized versions of the SaTScan algorithm are presented in this paper. Further, the MSCHD problem is modeled as an optimization problem and three improved variants of Particle Swarm Optimizer (PSO) namely I-PSO, HCL-PSO and Ensemble PSO are applied to detect the most significant hotspots. The comparative and sensitivity analysis are performed using synthetic datasets. The comparative analysis of SaTScan and the presented PSO based schemes is made in terms of the quality of identified hotspots and the computational time. Results reveal that the PSO based schemes (I-PSO, HCL-PSO, and Ensemble PSO) are promising and far efficient than randomized and traditional SaTScan algorithms. Further, for the datasets containing only a single hotspot, the performance of all PSO based schemes is at par with each other. However, for datasets with more than one hotspot in the study area, HCL-PSO has lower average rank than I-PSO and Ensemble PSO schemes and hence seems more promising for MSCHD. The superiority of HCL-PSO based hotspot detection is also validated using Friedman Test. Finally, the presented schemes are applied to the case study of Chicago city for identification of different types of crime hotspots.