FEATURE STORY — Mar./Apr. 2006

Inspector Gadget

Muscarello powered the Classification System for Serial Criminal Patterns with the Kohonen neural network, a self-training artificial intelligence program that, unlike most criminal analysis systems, adjusts to changes in criminal activity. “Criminals come and go,” he says. “We need a system that works dynamically.”

Serial criminals can hide—but
they can’t run from Tom Muscarello’s criminal analysis system

By Rachel Parker

In the late 1970s, Tom Muscarello ’71 LAS, PHD’93 ENG, was on the hunt for criminals. Peering over thick stacks of green bar paper saturated with numbers, the then-Medicare fraud investigator searched for data patterns that would reveal illegal activity. “I kept thinking there had to be a better way,” Muscarello says, sitting in his office at DePaul University’s School of Computer Science, where he serves as associate professor. “I wanted to create a tool that would help search through massive amounts of data.”

Nearly 30 years later, Muscarello has done it: His Classification System for Serial Criminal Patterns, or CSSCP, is capable of weeding through thousands of criminal cases and pinpointing patterns in variables such as weapons used and vehicles involved in crimes like robberies, rapes or murders. Developed in collaboration with Kamal Dahbur, DePaul professor and researcher, and Chuck Padgurskis, former director of the Chicago Police Department’s information systems, CSSCP will allow detectives to use patterns to track down serial criminals at a faster rate than if they had searched for patterns independently.

CPD will be the first to implement CSSCP, following field tests and system integration of the department’s recently upgraded “data warehouse”—possibly as soon as this year. However, CSSCP isn’t a turnkey software system. It must be configured on a case-by-case basis for each law enforcement agency’s data system.

CSSCP uses an artificial intelligence program called the Kohonen neural network. Unlike other system networks, the Kohonen neural network enables CSSCP to “train” itself and search for patterns 24 hours a day, without the prompt of a human operator.

The patterns are screened to ensure that they meet criteria, such as crime logistics. “If the system shows that an individual committed a robbery at 1 a.m. at the corner of Howard and Western and one at 1:12 a.m. down at 59th and Pulaski, that’s physically impossible [because they’re 20 miles apart],” Muscarello explains. “So the system uses a couple of different technologies—each of which is good at a different thing—to search for these kinds of things.” Eventually, written notes collected from detectives will be entered into CSSCP. “The text is where all the good stuff is,” he says. “If we can include that as part of the process, we’re going to have much better matches.”

As an undergraduate student at UIC, Muscarello took all of the computer programming courses available in the math department despite being a biology major. Following graduation, several odd jobs and a six-year stint as a Medicare investigator, he earned his master’s degree in computer science from DePaul, then a Ph.D. in electrical engineering and computer science from UIC.

Such training gives Muscarello confidence in CSSCP’s ability to fight crime. “It gives us an edge over criminals,” he says, his eyes steady behind a pair of wire-rimmed glasses. “If they want to play games, I don’t mind playing games.”