How To Catch a Criminal With Data

About seven years ago, researchers from the University of Memphis approached the city's police department with the idea that they might be able to detect patterns in local crime - geographic hot spots on the city's map and moments in time when they're most likely to flare up - if they could just have access to the department's crime data. Police departments produce reams of this stuff: arrest warrants, crime-scene reports, traffic citations, mug shots, dispatch transcripts and incident times. But that data has traditionally been painstaking to cross-reference, to mine for connections and even future trends.

The researchers ultimately turned the department onto a predictive software called SPSS, which had for years been used to crunch data in a host of disciplines not necessarily connected to crime. The department launched a pilot program with it to analyze trends, as part of a strategy of fighting crime by real-time data-mining.

"It brought about some resistance from some of the station command staff because, whereas the crime analysts had been doing a certain thing, now we're going to completely disturb the normalcy of what's gone on," says John F. Williams, the police department's crime analysis manager. "We're going to create an entirely brand new methodology in our approach to reducing crime in the Memphis area."

... Comparing the period of March 1-13 this year to the same stretch of 2006, the year in which Memphis really got this program underway, serious crime (the homicides, robberies, rapes, vehicle thefts, etc.) fell by 31.2 percent in the city. And it's not just that crime dropped, or that officers can now hand over stronger cases to prosecutors.

"The resources that we have now," Williams says, "will allow us to pretty much solve a crime far faster than what we had in the past."

... On a computer monitor, Knisley had pulled up a program called COPLINK, which sucks into one massive database all that disjointed information that was once scribbled down by hand.

... Knisley starts to plug this information into his system, as the officer actually did at the scene of the crime. First the software spits out 29 names. Then Knisley asks for a second male associate, and a two-door red car. Each piece of information is connected to myriad others: convicted criminals are linked to the names of every accomplice and victim known in the system, every car and street address they've ever been associated with, every alias they've gone by and arrest warrant they've been served. IBM has even been working on software that can reconcile typos and misspelled identities, correctly pegging Chriss Knisley, Chris Knisley and Chris Knisely as the same man.

Eventually, the system comes up with a suspect with 53 prior arrests - several for child abductions. In the real-life scenario, dispatches were sent out for all of the vehicles associated with his name, and the kidnapped child was recovered just half an hour after her abduction.

The software here was relying only on data already contained within the police department (this does suggest that a first-time offender would have been much harder to catch). But that data has never been this useful before.

Read more.

See also a more technical look in Communications of ACM, March 2012: Policing the Future

The Los Angeles Police Department utilizes three years of crime data to generate daily probability reports about when and where crimes are most likely to occur.