Software identifies ‘missing link’

New crime-linkage software has been unveiled in the US that can help police identify related crimes. It will enable analysts to more quickly and easily sift through often large volumes of crime data to accurately discover patterns that could indicate a series of linked crimes, such as burglaries.

Michael Porter, assistant professor of statistics at the University of Alabama, said establishing that a set of crimes is attributable to a common offender or set of offenders “is a critical first step to bringing an end to the crime spree and apprehending the criminal or criminals”.

He worked with Brian Reich, associate professor of statistics at North Carolina State University, to develop the crime linkage statistical model, that was presented at the 2015 Joint Statistical Meetings (JSM 2015) in Seattle.

Crime linkage helps identify a crime series – a group of crimes committed by the same person or group of people.

Mr Porter said currently police investigators manually sift through hundreds of cases and use their judgment to determine whether a series of crimes is linked. Even while using query-enabled databases it is a complicated task, requiring the crime analyst to make complex comparisons of the similarity and distinctiveness of crime scene characteristics, the criminal’s behaviors and any eyewitness descriptions.

The crime linkage model complements an analyst’s experience and knowledge by enabling the investigator to pool information from various crime scenes, strengthening the case against a serial offender and aiding in profiling the criminal.

“The model offers the potential to put linkage analysis on a more reliable and scientific basis, increases standardisation, reduces the workload of police investigators and improves the prospect of using linkage analysis as evidence in legal proceedings,” explained Mr Porter.

He said it could also predict the likely location of the next event in an offender’s crime series, an important aspect of tactical crime analysis. It does this by weighting more recent events stronger to capture changes in the criminal’s behavior – namely site-selection and crime-scene behaviors.

Mr Porter said in the course of planning and carrying out a crime and in response to the situations encountered, an offender “will make a series of decisions – perhaps unknowingly – resulting in a unique set of behaviors, commonly known as modus operandi”. Using statistical learning and data-mining algorithms, the model uses the measurable results of these decisions, recorded in a criminal incident database, to determine if crimes are linked.

Mr Porter and Mr Reich used 2001/06 data on 10,670 burglaries in Baltimore County, Maryland, to build and test the efficiency of their crime-fighting model. The results show the model was highly effective in identifying crime series, including the following results:

•The model captured 85 per cent of links with only five per cent false positives;

•The model has a precision of 91 per cent, indicating 91 per cent of the 100 highest-ranked pairs are true linkages;

•It identified 74 per cent to 89 per cent of true additional crimes from the series in a list of the top 50 crimes. When two or three crimes were already in the series, the model’s performance improved to 86 per cent to 91 per cent;

•A typical breaking-and-entering crime series was committed within 8.3sq miles and a period of 266 days; and

•Analysis of 590 crime series of four or more crimes indicates performance of next-event prediction varies greatly. If 10sq km is monitored for the next breaking-and-entering crime in a series, the model can detect an average of 65 per cent of the crimes.

Data from the FBI (Federal Bureau of Investigation) estimates there were nearly 2.2 million burglaries in the country in 2010 and arrests were made in only 12.4 per cent of the cases. The effectiveness of the model proves it is a promising crime-fighting tool that will be helpful to police investigators as they work to solve more burglaries, said Mr Porter.

To encourage use of the m