ICIAC 2012 Seminar stream 4B

Abstracts and slides

Crime linkage and investigative analysis (G)

Linking stranger rapes using fuzzy clustering

Don Casey and Phillip Burrell, London South Bank University

Key words: crime linkage, stranger rape, decision support, fuzzy clustering
Slides: Yet to be supplied by presenter        

This research investigates the possibility of using Artificial Intelligence (A.I) techniques to construct a computerised decision support system for use by crime analysts in finding links between stranger rapes.  The need for such a system to assist analysts in detecting serial offending was first identified by Grubin, et al in Linking serious sexual assaults through behaviour (2001) and has been repeated elsewhere in the literature. More recently a report (2012) by Her Majesty’s Inspectorate of Constabulary that examined rape investigation forcefully underlines the necessity of identifying series of stranger rape in its first recommendation: We recommend that forces should initially consider every ‘stranger’ rape to be part of a pattern of serial offending.

The aim is to reduce the area of search for linked offences by making a ‘first cut’ of the dataset of stranger rape that ranks offences by how likely they are to be linked to the index crime; and effectively directs the analyst’s attention to those that offer the greatest probability of linkage.

A number of A.I. methods have been used for decision support in crime analysis, but it was believed that Fuzzy Set Theory, which is a well-established theoretical approach, was particularly suited to this area because of its capability of dealing with concepts that are still imprecise and where the relationships between variables are not well understood.

In order to facilitate this work the Serious Crime Analysis Section of the N.P.I.A. made available a set of 112 linked and 22 unlinked stranger rapes from its ViCLAS database. The linked crimes comprised 40 series offences which were divided into development and test sets of 83 (29 series) and 29 (11 series) crimes respectively. The original dimensions of criminal behaviour used in the Grubin research: Control, Sex and Escape were taken as a starting point to describe each offence. Forty three variables were found that corresponded to those used by Grubin; and were distributed across the dimensions as Control (18), Sex (14) and Escape (11).

A fuzzy membership function was devised to express the degree to which each dimension was present in offences and the c-means fuzzy clustering algorithm used to associate the rapes in clusters.

The number of dimensions and clusters were varied and the degree of ‘fuzziness’ adjusted during testing. Each crime was compared with every other crime in terms of similarity in both development and test datasets and the rank distance found to be significantly lower between linked as compared to unlinked offences. This was also true when the 22 unlinked crimes were introduced. Results in 3 dimensions were particularly promising in the test dataset with linked crimes likely to be found within a third to a quarter of the expected search area. As serial offences consistently cluster much more closely to each other than could be expected by chance this suggests that the approach is a credible basis for a decision support system that could act as an initial screening mechanism in crime linkage analysis.

Catching a knife-point robber with geographical profiling

Christine Leist, Metropolitan Police

Key words: Geographical profiling, knife crime, robbery, investigative analysis, spatial-temporal analysis
Slides: Leist

This presentation is a practitioner’s case study of an unusual knife-point robbery series in London and the geographical profiling analysis supporting the investigation.

Robbery series can be particularly difficult to investigate. Often there is a lack of forensic evidence and it can be hard to identify links within volumes of offences with often indistinctive features. In the series in question the usual investigative methods had been fully exhausted and did not lead to the identification of the offender.

The presentation demonstrates how during the course of the analysis theoretical principles were translated into hands-on operational recommendations. It is explained how the geographical analysis assisted significantly with the linking of the series, even overwriting modus operandi as a linking feature since the offender behaviour was inconsistent. It explains the rationale of the analysis and how it complemented the investigation by providing a new approach and new possibilities for intervention.

Spatio-temporal considerations are outlined in detail and the presentation illustrates what conclusions were drawn about the offender’s spatial awareness and local knowledge. It is further explained how the general offender type was determined and what tactical recommendations were made as a consequence.

Finally a surprising twist reveals how the geographical analysis did solve the case but not without the creativity and dedication of experienced investigators. 

Page last modified on 30 jan 13 17:08