A Weapon to Fight Crime? Try Math
By Sean Brenner, Illustrations by Martin Short
Published Apr 1, 2012 8:00 AM
The best way to fight crime is to keep it from happening in the first place. And while they don't wield guns or carry badges, three UCLA faculty members are helping the Los Angeles Police Department do just that.
The scholars bring cold, hard science to predicting where certain crimes are most likely to be committed, a practice that, until now, relied largely on cops' experience and intuition. In development for six years, the "predictive policing" software developed by Anthropology Professor Jeffrey Brantingham and math professors Andrea Bertozzi and Martin Short, with key contributions from a few dozen undergraduates, grad students and doctoral candidates, helps cops identify potential crime hotspots and stop illegal activity before it takes place.
In its initial test, the software did a great job of befuddling bad guys. During one five-week period last fall, police recorded about 100 fewer burglaries and motor vehicle thefts in the city's Foothill section—which comprises Pacoima, Tujunga and Sun Valley and is home to about 250,000 people—than they had during the same timeframe in 2010.
At the heart of the work is the premise that certain crimes—home invasion, burglary and grand theft auto are prime examples—are more likely to occur in rapid succession and close proximity to one another, which Brantingham calls a "self-exciting process." Working from that hypothesis, the UCLA team set out to write mathematical equations that use data on recent criminal activity to predict new crimes.
"The LAPD already had a pretty sophisticated system for tracking crime," Bertozzi explains. "But once you get the data together, a lot of what takes place is eyeballing crime maps, discussing it and planning what to do next. Our goal was to not only better quantify the data, but to understand trends in the data and to make better use of it. How could we better represent data in a way that's systematically correct?" The program generates color-coded "heat maps" that indicate the "highest probability areas for where crime will occur that day," Short says. "This doesn't replace the intuition of people who have been working on the street for years, but it brings them more precision and accuracy, and it doesn't take years of experience to use."
Captain Sean Malinowski is the LAPD's point person on the initiative. "At UCLA, you have anthropologists, criminologists and math scholars—some of the best in the world," he says. "That has been a huge boon to us. And it's significant when you have 15 to 20 fewer people per week coming home to see their homes or cars broken into."
Malinowski says that if the pilot program is successful, the model could be rolled out to the rest of the San Fernando Valley or even citywide. And the UC LA team, which has formed a company to market the software, has designs on partnerships with other law enforcement agencies.
"The most striking thing about the partnership with UCLA is that while the professors are getting experience and data out of it, they're doing it out of a sense of public service," Malinowski says. "These guys get very excited when their work prevents someone from being victimized."