NCJ Number
204581
Journal
International Journal of Police Science & Management Volume: 5 Issue: 4 Dated: Winter 2003 Pages: 265-276
Date Published
2003
Length
12 pages
Annotation
This study assessed the merit of data-mining techniques for crime analysis, with a focus on domestic and commercial burglaries in West Midlands, England.
Abstract
The study applied three data-mining techniques -- the multi-layer perceptron (MLP), radial basis function (RBF), and the Kohonen self-organizing map (SOM) -- to the building descriptions, modus operandi (MO), and temporal and special attributes of domestic and commercial burglaries attributed to a network of offenders. The authors discuss the benefits of extracting a formal structure from a database field that contains unstructured MO text and then using this structure in the data-mining process. Also explained are the stages of data selection, coding, and cleaning. Two police sergeants from the West Midlands force, who were not part of the research team, independently validated the results achieved by this process. The burglary database used in the study contained 23,382 recorded offenses that occurred between January 1997 and February 11, 2001. A total of 4,159 offenses had been traced to a variety of offenders who had committed 17.79 percent of the total number of burglaries. This included 214 burglaries attributed to a known network of offenders (primary network) who represented 0.92 percent of the total burglaries and 5.15 percent of all cleared burglaries. The primary network offenders' MO was to target shops and gas stations. The data were encoded from text by a team of specialists following a well-defined protocol, and they were analyzed with the MLP, RBF, and SOM techniques contained within the data-mining workbench of the commercial data-mining package SPSS/Clementine. Within minutes, 3 months of unsolved crimes were analyzed through the Clementine stream, producing a list of burglaries that might be attributed to the primary network of offenders. The analysis by the two police sergeants determined that 85 percent of the selected burglaries could be attributed to the primary network of offenders. The production of a manual list of such burglaries would have taken between one and a half to 2 hours and would have yielded between 5 percent to 10 percent accuracy. Study limitations and further research are identified. 4 figures, 2 tables, and 19 references