Predictive Models For Public Safety Using Social Network Analysis

Download or read book Predictive Models for Public Safety Using Social Network Analysis written by Mohammad Ali Tayebi and published by . This book was released on 2015 with total page 133 pages. Available in PDF, EPUB and Kindle. Book excerpt: Crime reduction and prevention is the major concern of law enforcement agencies in order to increase public safety, reduce the crime costs to society and protect the personal integrity and property of citizens. Along with big data analytics, predictive policing which is a new paradigm for crime analysis has been emerging. An important task in predictive policing is analyzing the relationships between offenders to fully understand the criminal collaboration patterns. Law enforcement agencies have long realized the importance of analyzing co-offending networks, networks of offenders who have committed crimes together, for designing prevention and intervention strategies. Despite the importance of co-offending network analysis for public safety, computational methods for analyzing large-scale criminal networks are rather premature. In this research, we study co-offending network analysis as effective tool assisting predictive policing. We start with a formal representation of crime data and co-offending networks to bridge the conceptual gap between abstract crime data level and co-offending network mining. To gain a better understanding of co-offending networks we thoroughly study their structural properties. Specifically, how centrality measures can be employed to identify key players of co-offending networks to disrupt these networks. Then, we propose an algorithmic solution for detecting organized crime groups from a social network analysis perspective. We explore predicting criminal collaborations from two angles. First, given partial information about offenders involved in a crime incident, our proposed approach assists in investigating the most probable suspects in that incident. Second, we propose a supervised learning framework for co-offence prediction. Finally, we propose a random walk based method for crime location prediction which is personalized for every offender by using spatial information about offenders and the co-offending networks structure. The efficacy of the proposed methods are experimentally evaluated using a large crime dataset representing five years of police arrest-data for the regions of the Province of British Columbia which are policed by the RCMP, and compared to other existing methods.