Evaluation of Unsupervised Classification on Police Patrol Zone Design Problem

Police patrols are one of the effective ways to respond a recent incident or to prevent a likely crime. However, due to limited number of available officers, not all streets can be patrolled continuously. To minimize the potential crimes, available forces should be systematically assigned to their zones, where an incident is most likely to happen. In this study, we propose a new technique that automatically determines the patrol zones and shifts, then assigns available officers to the zones. Our technique utilizes the previous crime data of the region and employs unsupervised classification to automatically identify zones. Then, it estimates a crime weight for each zone based on several factors, such as the probability and type of an incident. Finally, it assigns officers to the zones by prioritizing crime weight estimations. As a case study, we tested our system on the city of Montgomery, AL. We used the crime dataset including all reported 19687 incidents in 2017. We have built our scheduling system using 11 months of data and then tested the system using the incidents happened in the last month. Test results show that, with the proposed technique, scheduled officers are most likely to present nearby the incident before it happens. We measured 7 minutes of response time on the average, which is lower than the national average.