PolRoute-DS: a Crime Dataset for Optimization-based Police Patrol Routing

Abstract

It is a well-known fact that criminality is an open, yet important, issue in most urban centers worldwide. Especially in Brazil, creating solutions to reduce crime rates is a top priority. To reduce crime rates, many cities are adopting predictive policing techniques. Predictive policing techniques are highly based on extracting valuable knowledge from a massive dataset that contains information about times, locations, and types of past crimes. The extracted knowledge is then used to provide insights to police departments to define where the police must maintain its presence. These datasets may also be used for a critical predictive policing task: defining where police patrols should patrol. Such patrols are commonly defined to cover a series of crime hot spots (areas that present high criminality levels) and have some restrictions to be considered (number of available police officers, cars, etc). Thus, defining the route for each police vehicle is a complex optimization problem, since in most cases, there are many hot spots and the existing resources are scarce, i.e., the amount of vehicles and police available is much smaller than necessary. Unfortunately, high-quality crime rates data are not easy to obtain. Aiming to tackle this problem, this article proposes the PolRoute-DS dataset, a dataset designed to foster the development and evaluation of police routing approaches in large urban centers. The PolRoute-DS combines the spatial structure of the city of interest (in the context of this article, the city of São Paulo) represented as a connected and directed graph of street segments with criminal data obtained from public sources. PolRoute-DS is available for public use under the Creative Commons By Attribution 4.0 International license (CSV and PostgreSQL versions) and can be downloaded at https://osf.io/mxrgu/.