When George Mohler was an undergraduate mathematics student at Indiana University, he had no idea that his work would eventually have the potential to shape crime prevention and incident response for major metropolitan police departments.
Mohler's Ph.D. studies in the University of California system steered him into predictive modeling and data analytics. His first assignment was in Los Angeles -- no small task.
"After grad school I got an offer from UCLA to work on a NSF funded project that was a collaboration between applied mathematicians, social scientists, and the LAPD," said Mohler, now an associate professor in IUPUI's Department of Computer and Information Science. "It sounded interesting and something different: My graduate work focused on complex fluids. I worked there for several years and we had some success with a prototype predictive tool we developed to direct patrols. We received media attention on several papers we wrote, which led to a number of inquiries from police departments for the software we had written."
Mohler is currently collaborating with Jeremy Carter, IUPUI associate professor of criminal justice, and Rajeev Raje, IUPUI professor of computer science, on a project crunching huge data sets of "social harm" -- traffic accidents, medical emergencies and drug overdoses to burglaries, car thefts and homicides. The team is researching whether the property crime models deemed a success in Los Angeles can be extended to handle this wide array of social harm in Indianapolis.
After Mohler solidifies the model for Indianapolis' perimeters, the work will eventually appear as an app for mobile devices used by IMPD, with spring 2019 targeted as the launch date for a field trial. The goals of the implementation are to identify trends to help set up more-efficient patrols and quicker responses to emergency calls.
We asked Mohler to dive deeper into how his work will affect our city.
Q: How did you gather the initial data that fuels this software?
A: Jeremy Carter came up with a way to quantify the average social harm of each event type. If you look at a burglary, there are different kinds of costs: cost to law enforcement, cost from investigating the burglary. When somebody's charged, there are court costs. There are insurance costs for the victims. You try to add up all of these societal costs and quantify the burglary on average. We're taking those sorts of estimates and pairing them with predictive models of where and when crime and other types of events are likely to occur. Those are based on point processes, which is a type of model that is good at modeling space, time and event patterns.
We try to allocate resources that hopefully can prevent the largest amount of social harm cost to the city.
Q: How does the software modify with changing crime trends in a city?
A: We take more of a machine learning approach. Your city may have an increase in a certain type of event. It could be homicides, it could be drug overdoses, and it's not going to be all over the city. It's going to be in certain areas at certain times of the day.
We try to design flexible algorithms that can detect those patterns as they emerge. Then, hopefully, the police have a software application that can allow them to easily divert patrols or other resources to those areas.
If I'm an officer using the app, I log in each day for my shift, and these are the areas I should patrol when I'm not on a call to service.
Q: How did your experience in Los Angeles shape your IUPUI research?
A: In Los Angeles, our randomized controlled trials of predictive policing focused on property crime. Police were using predictive analytics to allocate patrol resources to reduce property crime in several divisions in Los Angeles. This research builds on that work where now we're expanding a few things: the types of events -- not just property crime, but traffic accidents, medical emergencies and other social harms and looping in more stakeholders – IMPD, EMS, neighborhood watch groups, etc. will have access to the application.
Q: What's next for your research?
A: I have another grant on algorithms for threat detection. The IMPD project is for more routine activities. They happen every day and they aren't really anomalies. In space-time-event data, how do you predict anomalous patterns? It could be terrorism or natural disasters and people's responses to those, with applications to where resources are needed in real-time.