12.3 Crime Clusters

One of the most notable characteristics of crime is that it tends to organize itself in distinct clusters. Criminal acts do not extend evenly over space, and they are not constant over time; there are variations and patterns in space and time. As far back as the 19th century, statisticians have closely studied the differences in crime across communities. Today, clustering and its implications still play a central role in the study of crime. Clustering occurs in both temporal and cross-sectional data and in both individual and aggregate analyses. Criminologists often perform different type of clustering analyzes, such as individual temporal, aggregate cross-sectional, individual cross-sectional and aggregate temporal.

As stated earlier, yet another method of developing clusters is via the use of data mining to find anomalies, such as discovering unexpected hidden associations between a class of crimes and perpetrators' MO. So far most of the criminal mapping has been human-driven, where hypotheses are performed on the data in search of patterns and clusters. However, data mining, as we have found out, can find unexpected patterns and clusters organized by the data itself, using special types of neural networks. Innovative criminal investigators in the United Kingdom and the United States have begun to use these types of data mining techniques in their quest to solve crimes and gain insight into the crime maps.

As with many other police department, the West Midlands police department in the United Kingdom is faced with shrinking resources, few leads, and aging cases. Investigators find that these challenges can limit the cases they can investigate. High-volume cases without definite leads, such as house burglaries and vehicle theft, that lack clear evidence are often filed away until new evidence is found. However, each West Midlands electronic case file contains physical descriptions of the thieves, as well as their MO. While many cases lacking evidence were previously filed away, the department is now reexamining them with a new type of weapon: data mining.

Inspector Rick Adderley is using two (SOM) Kohonen neural networks to cluster similar physical descriptions and MOs. He then combines clusters to see whether groups of similar physical descriptions coincide with groups of similar MOs. If he finds a good match and perpetrators are known for one or more of the offenses, it is possible that the unsolved cases were committed by the same individuals. Adderley's analytical team further investigates the clusters, using statistical methods to verify the importance of these similarities. If clusters indicate the same criminal might be at work, the department is likely to reopen and investigate the other crimes. Or, if the criminal is unknown but a large cluster indicates the same offender, the leads from these cases can be combined and the case reprioritized.

Adderley is also investigating the behavior of prolific repeat offenders with the goal of identifying crimes that seem to fit their behavioral pattern. He constructs a model using data from cases in which the offender is known, and then applies it to a database of unsolved crimes. Such models can be an invaluable aid in linking known criminals to specific crimes quickly. The following case studies are presented in their original versions in which SOMs are used to construct clusters of criminal data. These analyses signal the advent of a new type of forensic data mining techniques. The following case study is provided in its original version; the author would like to thank Inspector Rick Adderley for his valuable assistance in providing the paper on his innovative work with SOMs.