Hi-Tech Crime Predictor
More than a decade of extensive crime data collection matched with new technology may soon allow police to predict the number and type of crimes that will occur in a given neighborhood one month in advance.
Researchers at Carnegie Mellon University plan to release software later this summer that, according to computer models, can predict the number and types of crimes that will occur within a 10-block area with a 20 percent error rate.
"This is the next generation of crime mapping," said Wilpen Gorr, a Carnegie Mellon professor of public policy and management information systems.
A small team of researchers are running final tests on the project funded by the National Institute of Justice, the research wing of the Department of Justice.
Researchers used crime data from Pittsburgh and Rochester, N.Y., collected between 1990 and 2001.
With models similar to those used in macroeconomics to warn of recession or inflation, Gorr said researchers matched criminal reports, 911 calls on shots fired, and census data along with a mishmash of yellow page listings and seasonal variations to predict crime numbers.
After running 36,000 separate forecasts for the month ahead based on a decade's worth of crime data, researchers successfully predicted how many burglaries, arsons, aggravated assaults and car thefts and other crimes would take place in the next month in a 10-block area, Gorr said.
"It's surprising that it worked at all and the 20 percent error area was impressive," Gorr said. The new system also foretold crimes that would not likely have been anticipated using traditional crime mapping, he said.
"I think more impressive is that 30 to 40 percent of the crimes committed could be seen as a real big surprise, something you would not have expected in that area at that time."
Murder and some other major crimes were not included because they are scarce, relative to offenses such as larceny, Gorr said.
There are a number of research projects underway that have the same goal, but use different data.
Ned Levine, a Houston-based transportation specialist, is using travel patterns of criminals (where criminals are arrested and a home address listed on police reports), the travel patterns of victims, plus existing crime data to predict where the two will intersect and when.
"Certainly, these systems are not perfect, but GIS (geographic information systems) and other spatial technology are so far advanced that we are a world away from the crime mapping of 10 years ago."
Levine expects to develop a prototype by fall, but a working model is likely years away, he said.
Police departments have used crime mapping for years with technology advancing from pin maps to multilayered, computer-generated illustrations.
Most beat cops have a general idea of what crimes to expect based on their knowledge of a neighborhood and the time of year. Crime mapping has allowed for an even more comprehensive reaction to clusters of crime, said Philip McGuire, assistant commissioner of programs and policies for the New York Police Department.
Any advance in mapping technology would be welcome, he said.
"An officer might know some of this stuff in his gut, but anytime you can nail that down, it allows you to take that information and build it up to the macro level, where you're talking about the entire city, the county or state," McGuire said. "If it's a more accepted position because it's science, not the gut feeling of a street officer, that is going to be heard by people making decisions for the department."