A Debiasing Technique for Place-based Algorithmic Patrol Management

As police departments have come to rely on algorithmic patrol management systems to assign patrols, community groups and academics have raised concerns about demographic bias in the data used to train these systems. This paper introduces a technique for eliminating demographically correlated features from a place-based algorithmic patrol management system. We apply our technique to the real-world algorithmic patrol management system ResourceRouter, developed by SoundThinking. After applying the technique, we test the system's accuracy against historical crime incident data in four cities across several crime categories. Results suggest that demographically correlated features can be removed from the system without reducing accuracy.