OREGON STATE UNIVERSITY

Researcher helps decode complexity in law enforcement data

10/13/2009

Incident Type: Drug investigation

Incident Date & Time: 1/28/2008 7:22 p.m.

Address: 700 block of South Park Street

Weapon Found: Beretta 92

Arrested/Suspect(s): Suspect is described as a white male, early 30s, 5’10” – 6’0”, medium build, dark hair. He was arrested for possession of a firearm, possession with intent to deliver cocaine.

CORVALLIS, Ore – Crime statistics like those above appear frequently on popular television dramas, where a swarm of detectives, crime scene investigators, and prosecutors work on one case involving a single key suspect and his MO.

But for Byron Marshall, an assistant professor in information management at Oregon State University, the science of crime is more complex than TV dramas might lead you to believe.

Marshall wants to make law enforcement easier – not by reducing criminal activity down to a handful of facts, but by finding ways to organize, analyze and make sense out of very large and diverse collections of law enforcement data.

“Link charts” are an essential tool in the law enforcement domain. They depict individuals and relationships discovered during an investigation and are used to focus investigations, communicate within and between agencies, and present data in court. In Marshall’s recent research, which will be published in a forthcoming issue of the Journal of the American Society for Information Science and Technology, he introduces an algorithm that helps identify important associations among a subset of individuals out of large data sets of criminal incidents to help investigators create better link charts faster.

The data set Marshall used in his study includes details from 5.2 million incidents and 2.2 million people, drawn from incidents recorded by the Tucson Police Department and the Pima County Sheriff’s Department in Arizona. Two investigations provided the experimental test data. One case involved fraud and methamphetamine trafficking and the other focused on a single criminal investigation.

Link charts are currently produced manually. For a major case, it can take an experienced crime analyst several weeks to create a link chart. Marshall’s algorithm promises a way to streamline the process.

The challenge for Marshall was to create a program that could take advantage of the power of computing without sacrificing the intuitive element humans contribute — an element that can be essential to solving crime.

“Humans can do things that are more subtle and intricate than basic mathematical extraction techniques alone can accomplish,” said Marshall. “So the driving force behind my research is to find ways to do searches better and faster by starting with a flexible computer algorithm that allows the addition of nonstandard information and notions from the user during the course of the analysis.”

Marshall’s computation starts with the target individual and looks at all other individuals in the criminal activities database to find connections to the target individual. “We begin with one person and we ‘pass importance’ out to everyone else,” said Marshall. “Certain people rise to the top either because they are connected to the target or because they are involved in particularly noteworthy activity as defined by the detectives.”

Marshall’s research benefits criminal investigators in several ways. In the past, some major criminal case investigators created profiles and then searched through data for individuals who fit the profiles. This approach may run counter to fairness ideals such as due process and probable cause. Marshall’s approach, on the other hand, starts with a specific target instead of “fishing” for suspects, thus reducing the potential for civil rights lawsuits.

By beginning with a target and spreading (or “flooding”) to individuals connected to the target through documented associations and activities and by employing heuristics normally used by investigators, Marshall’s technique produces outcomes that are useful for law enforcement tasks and appropriate for criminal justice purposes.

Although his computer application addresses a real-world investigational task – link chart creation – the same technique can be expanded, for example,  to aid biologists looking at the vast amount of literature in medical journals or business people doing patent searches. He hopes to expand his research stream by broadening the technique to other applications in the areas of bioinformatics (for example, literature searches for related studies) and K-12 education (for identifying lesson plans that meet specific educational guidelines).

“My goal is to help the experts guess better,” Marshall said.