Game Theory Can Help Predict Crime

illustration of policeman and criminal


About a decade ago, the hit movie Minority Report featured a police force that could predict crimes and swoop in before they happened. That kind of crime fighting may not be far off if a team headed by Eugene Vorobeychik, assistant professor of computer science and computer engineering, has its way.

While the movie cops relied on psychics to determine potential perpetrators, Vorobeychik and his collaborators use data, computing and analysis.

The transinstitutional team of Vorobeychik; Kenneth Pence, BS’77, MS’03, PhD’05, associate professor of the practice of engineering management; and Paul Speer, professor of human and organizational development, uses game theory and big data to optimize policing.

The researchers use crime data collected by the Metropolitan Nashville Police Department. They also use data from property assessors and even weather information, plus some valuable insider insight: Pence is a graduate of the FBI National Academy and a 31-year Nashville police veteran.

Three main classes of methods are used by police to determine force concentration in order to prevent or quickly respond to crime. The first, arguably most widely in use, uses crime data to determine hot spots of criminal activity where police patrols are subsequently deployed.

The second approach, termed “terrain risk-based modeling,” uses terrain layers (such as locations of bars, convenience stores, pawn shops and the like) that are useful at predicting crime incidence. This approach, again, can be used to predict criminal activity, and police patrols can then be deployed efficiently to anticipate and prevent crime.

“A relatively recent third alternative uses a mathematical crime diffusion model in which locations have differential attractiveness depending on environmental factors and past crime locations to predict crime incidence, as well as anticipate crime response to police presence,” Vorobeychik says.

The ultimate goal is to predict space–time crime incidence for crimes that have yet to occur.

Vorobeychik says the team is building on game theoretic security methods that already have been successfully deployed by the Los Angeles airport for canine patrols, the Federal Air Marshall Service for scheduling air marshals on planes, and the U.S. Coast Guard for scheduling boat patrols in New York Harbor.

“Since pinpointing crimes precisely is unlikely, we will instead produce a probability distribution—a ‘risk map’—of crime in time and space,” says Vorobeychik, who adds that they are “collating and crunching the data” now. The project is financed primarily by an Interdisciplinary Discovery Grant from Vanderbilt.

This probability distribution will depend on the following factors: past crime incidence (hot spotting), weather, location of businesses that tend to correlate with higher crime risk (bars, strip clubs, bus stops, check-cashing outlets, pawn shops, fast-food restaurants and liquor stores), as well as locations of known gang members and drug arrests (risk terrain modeling), and location of police over time and space.

Using game-theory modeling, the researchers can gauge the effectiveness of police patrols and expert opinions of police personnel to assess solutions proposed. Then they will experiment with different techniques in the field, determining the most effective model to be adopted by police.