Holistic Study of Crash Hotspot Identification Methods for Law Enforcement Resource Allocation

In recent years, as high-quality crash and other safety data has become available, law enforcement agencies have begun to implement data-driven approaches for traffic safety resource allocation and enforcement. Hotspot detection methods, in particular, have proven to significantly reduce crimes, and this concept has recently been extended to traffic crash prevention. The Wisconsin State Patrol in 2019 developed its own hotspot detection and analysis tool as part of its Predictive Analytics program that utilizes historic crash data to plan high visibility enforcement activities. This tool has been used successfully for several Predictive Analytics enforcement cycles by different agencies across multiple counties and regions in the state. Despite this success, there are known limitations in the hotspot detection capability as well as a desire to conduct a more thorough comparison to prevailing methods in order to better assess the quality of the tool. This paper examines several prevailing clustering algorithms based on different approaches, including centroid, density, hierarchical based and grid based methods, to investigate which approaches are most suitable for the law enforcement context and to suggest potential improvements to the Wisconsin algorithm. A comprehensive comparison of the prevailing methods against real-world data is presented along with conclusions. Overall, the density-based methods provided the best results. The Wisconsin Community Maps hotspot detection, which is based on a density approach, is comparable to the prevailing density methods such as DBSCAN and OPTICS, however none of the methods that were investigated provided significant advantages in terms of adapting to the variety of real-world scenarios in the Wisconsin crash data.