Design and Implementation of an Automated Anomaly Detection System for Crime

Anomaly detection in the law enforcement domain is very important because it gives police agencies a method for discerning the difference between normal fluctuations and fundamental changes in a specific crime distribution. Currently, crime analysts must monitor the data manually to detect these anomalies because an automated tool does not exist. This paper describes the analysis, design, and implementation of a system, entitled Sentinel, which directly addresses the anomaly detection problem. The Sentinel system is comprised of a robust anomaly detection back-end program built around a zero modified Poisson model and a user-friendly web interface. The back-end program constantly monitors user specified crime levels over time and automatically alerts users via the web interface when these crime levels change significantly. This tool has been completed and integrated into the Web-based criminal analysis toolkit (WebCAT), a criminal analysis program developed by the Systems Engineering Department at the University of Virginia. Although more long term comprehensive testing remains, initial results show that the tool provides a more accurate, time efficient method of detecting anomalies in criminal data sets than currently exists.