Tuesday, September 13, 2011

Science!

Applied mathematics:
The police department in Santa Cruz, California, has begun an experiment that uses a mathematical algorithm to predict when and where certain crimes will be committed, and puts police on the scene before they happen.

So far police have arrested five people using this technique of "predictive policing" and the rates of certain categories of crimes in the city have dropped significantly, perhaps as a result. The program has correctly predicted 40 percent of the crimes it was designed to monitor.

...The program comes from the field of applied mathematics or operations research, and the algorithm was developed by a 29-year-old mathematician at Santa Clara University.

Other mathematical techniques have been developed to predict crimes, most famously Compstat, used in the mid-90s by the New York City Police Department to track serious crimes, like those depicted in the Minority Report. The Santa Cruz program, which does not appear to have a name, concentrates of property crimes, such as car break-ins and burglaries.

The program was developed by George Mohler, an assistant professor of mathematics.

The algorithm he uses is based on computations used to predict aftershocks following a large earthquake.

The heart of the program is the belief that criminals often commit a second or third crime in the same location and the same time as a first successful crime. For example, if a burglar is successful breaking into a home at 2 p.m. in a certain neighborhood because no one is home, the criminal will use that experience to do it again to another house in the same neighborhood around the same time.

In the case of Santa Cruz, on California's central coast and home to a University of California campus, that would be about four days later.

The algorithm knows this because Mohler has fed eight years of data on crimes in Santa Cruz into the algorithm.

He first tested the notion in Southern California's San Fernando Valley with data from the Los Angeles Police Department. After a story about the project appeared in the Los Angeles Times, Zach Friend, a crime analyst at the Santa Cruz department contacted Mohler and gave him data from 2002-2010. Uniquely, the data is updated daily, something other programs like Compstat don't do.

As more data is slipped into the algorithm, the program is believed to get more accurate.

"The overall model is based on the belief that crime is not random," Friend said."So with enough data points you could predict where and when it will happen."

No comments:

Post a Comment