Criminal Activity Spaces and Crime Linkage Analysis: Towards a Suspect Prioritization System

Offender spatial behaviour has been utilized extensively by practitioners and researchers alike as a vector to study criminal behaviour. Psychologists and criminologists alike have been utilizing spatial behaviour as a means to describe crime patterns, which has resulted in several areas of specific research. Several prominent theories of criminal behaviour resulted from this work including routine activity theory and crime pattern theory. This thesis builds upon past research by reintroducing activity space as a means to study criminal spatial movement and the progression of criminality in general

First, this thesis addresses the lack of specific methodologies for a standard estimation process for criminal activity spaces. There is no consensus among the existent activity space literature as to how to calculate such spaces, which is further compounded by the unique challenges associated with working with crime data. As such, the current thesis introduces methods for estimating activity spaces as a single or composite geometric surface given the entirety of a given offenders criminal history. From this standardized activity space construct, general properties of spatial involvement (size, dispersion) were calculated as well as the proportion of offenders for whom the activity space for a given number of crimes included the next temporally sequential crime in their series.

Furthermore, this thesis demonstrates the utility of an activity space based framework in studying criminal spatial behaviour through direct example via crime linkage analysis. Activity spaces for all serial offenders (n = 997) within the available sample were calculated and used within a crime linkage exercise. Results indicate that such methods can be greatly improved by incorporating specific activity space based measures via two distinct studies: the first utilizing logistic regression models in accordance with past research; the second employing a more sophisticated machine learning technique of random forest modelling.

Finally, this thesis demonstrates how activity spaces can be used to extend crime linkage methods away from the traditional crime-focused approach to an offender-focused approach. Past studies of crime linkage analysis have used spatial proximity to assess strict crime-to-crime relationships, however results from a suspect prioritization simulation indicated that an activity space-based one-to-many comparison resulted in significantly improved suspect prioritization performance: traditional models correctly prioritized the correct offender into the top five suspects 24% of the time compared to 32% of the time for activity space-based approaches. This body of work serves to illustrate not just the suitability of the proposed method of modelling criminal activity spaces, but also the theoretical implications contained therein.