This Small Business Innovative Research (SBIR) Phase II project will further develop HunchLab -- software tools that leverage spatial statistics to enable police personnel to test their theories of criminality against data collected in the day-to-day activities of policing. The preceding Phase I project proved the feasibility of developing HunchLab as a set of innovative software tools that scour the historic data of a police department, search for geographic aberrations expected by the theories or 'hunches' put forth by crime analysts, and apply spatial statistics to confirm or deny the supposition. Preventing crime is a more sophisticated task than simply mapping incidents or arrests and deploying resources accordingly. The ability to detect and analyze changes in the geographic patterns of crime and disorder is an innovation in policing which holds the potential to enhance the organizational capacity of police departments across the country. This Phase II project will refine the application and build additional functionality, including alternate workflows for different user types, expanding the alert infrastructure, and building text mining capabilities.
The obvious sector that this product will impact is law enforcement at all levels of government. Additionally the successful outcome will impact federal law enforcement agencies and regional crime analysis consortia. There are roughly 250 municipalities with over 100,000 people in them, and these each have police departments that would find this system of use. The tools will be helping thousands of police officers do their jobs better every day. This efficiency will result in better policing, meaning that criminals will be caught more effectively. Criminals cause damage far in excess of the property and medical costs directly attributable to their activity. Perhaps more importantly, the research will form the basis for other products that operate in realms other than law enforcement. The algorithms and technologies developed in the Phase I prototype are transferable to other datasets that demonstrate similar point pattern processes - events with explicit spatial and temporal attributes. Our Phase I process demonstrated a substantial utility in domains other than law enforcement including fraud detection, real estate, sales and public health. The Phase II work plan includes testing with other data sets to refine that software should address these other markets.
The modern crime analyst uses digital mapping software to identify and display patterns and trends in crime. In a large city, this can mean sorting through millions of records generated by a police department in a given year. The HunchLab product developed under this NSF SBIR grant leveraged the latest geospatial data analysis, criminology and high performance computing techniques to enable law enforcement agencies to identify new geospatial and temporal crime patterns as they are emerging. The research and development work brought together the latest science from spatial statistics, geography, criminology and computer science to create a powerful data mining system that can enhance public safety by providing more timely intelligence to law enforcement agencies. HunchLab is capable of sifting through vast amounts of location-based events and automatically identifying areas in which recent events demonstrate statistically significant changes. The product is designed to help local and regional law enforcement agencies to more effectively combat crime through automated discovery of space-time event patterns and innovative display and visualization tools for evaluation of the patterns. During the course of the Phase II implementation, Azavea extended HunchLab from being solely focused as a "crime early warning" system to also include more generic crime analysis and visualization needs. Under the Phase IIB effort, the system was further extended to support risk forecasting and predictive analytics. Two different crime "forecasting" techniques were implemented. The Near Repeat technique uses the results of recent crime geography research that has demonstrated that in the few days and weeks following some crimes, there is an elevated risk of the same crime occurring nearby. We can now measure that risk and plot it on a map, enabling more targeted patrols. The second forecasting technique measures the degree to which crimes vary over the course of a year, a week and a day and measures the cyclicality as it relates to specific geographic areas. This seasonality, time of day and day of week component can be calculated and automatically applied to arrive at a forecast for a month or a year, against which law enforcement officials can measure their progress in reducing crime. The HunchLab software was developed using 15 years of crime data in Philadelphia but has already been deployed in Tacoma, Washington with several additional departments expressing interest in implementing the system.
