We’ve been keeping a secret for the past year. Toiling away on our crime modeling research, we have been building a system that can predict the risk of crime at geographic micro-places by incorporating many data sets into one, unified predictive model. This has led us to the next generation of our HunchLab product, a vision we recently unveiled as HunchLab 2.0.
Modeling crime to allocate police resources appropriately is not new, but using advanced machine learning and statistical modeling techniques to improve the accuracy of resource allocation is quite new in the law enforcement community. Underlying the new HunchLab is a crime forecasting system that balances the importance of many crime theories in accurately predicting risk across time and space.
For instance, picture yourself in charge of allocating police vehicles to patrol the 6th District. There are many factors that may impact your decision: recent crime events, long-term hot-spots of crime, the location of subway stops and bars, the day of the week, the season, the weather, or the Eagles game occurring that afternoon. By building a system that automatically determines what data sets best predict each type of crime, we can aid police commanders in making better decisions to ultimately improve public safety.
The predictive output of this new forecasting system and continuing advances in mobile technology and the elasticity of cloud hosting allowed us to re-imagine the way officers, analysts, and commanders could use predictive policing software. For example, HunchLab enables police commanders to allocate resources based upon the predicted societal impact of crime at different locations for each individual shift, weighting the importance of preventing different crimes appropriately. This data-driven process leads to resource allocations that accurately and fairly reflect societal priorities for public safety, unbiased by neighborhood affluence, race or ethnicity.
Additionally, the software assists analysts in promoting evidence-based interventions within these continually evolving focus areas by publishing both their analysis notes and crime-specific tactical responses to the field. Officers receive risk assessments for their current location and contextual information about focus areas through a location-based interface we call Sidekick. Displayed on a vehicle’s laptop, a smartphone, or a tablet, we use the officer’s GPS location to push relevant information to them as they move about the municipality.
We’ve been testing the predictive accuracy of the software with data sets from Philadelphia, Chicago and Lincoln, NE. We are currently working with police departments to test this new technology in the field and provide us feedback on this bold new vision for policing.
The early development of HunchLab was made possible in part thanks to funding from the National Science Foundation’s Small Business Innovation Research (SBIR) program (IIP-0637589, IIP-0750507).