Optimizing metrics in police routing algorithms

Abstract

A large part of the mission of state troopers is to prevent traffic accidents and to quickly respond to the accidents that do happen. However, driving about aimlessly during their shift is not efficient. Certain areas can be identified as "hotspots", places where crashes are known to frequently occur. It is advantageous to have officers target these critical locations during their patrol routes. Multiple officers taking similar routes is also inefficient. The number of officers patrolling is limited, and by keeping them spread out, response time to crashes can be decreased. The purpose of the Turn programming language is to create efficient routes daily, but with a degree of randomness to prevent the routes from becoming predictable. At its core is a graph representing the roads of Alabama, with vertices at each milepost and intersection. Turn programs utilize set reduction functions to choose what vertices officers should patrol. Depending on what functions the programmer uses and the order they are used, the route may be different to reflect the changing priorities. A Turn program's worth is measured by a number of metrics, such as how many hotspots were covered each day, how long those hotspots were patrolled, and time taken to respond to crashes in the simulation. Additionally, a program is worthless if the routes it creates are not realistic. In this paper, we present an analysis of various Turn programs, explain how they affect the metrics, and show a program that strikes a balance between them.