Supporting the use of web-based crime mapping in cooperative policing by dynamic pattern analysis

Abstract

This thesis reports on a map-centric spatial-temporal crime pattern analysis system -Dynamic Pattern Analysis system (abbreviated as DPA system). The term "dynamic pattern" does not refer to a particular type of spatial-temporal pattern. We use the term "dynamic" to reflect the dynamic and diverse nature of spatial-temporal crime patterns. The proposed DPA system consists of four modules: (1) Basic tendency analysis provides a general view of the distribution of crimes over space and time, (2) Spatial relation analysis identifies spatial relationships of crime patterns at different time points, (3) Cyclic Signature clustering makes use of similarities of calendric patterns to find hotspots with similar features, and (4) the Dynamic Relation model is a data-driven evaluation approach for analysing the interactions between spatial-temporal crime patterns and Web-Based Crime Map. Four types of dynamic relation (HI, HD, LI and LD), referring to different situations, can be identified by the Dynamic Relation model. The different types of dynamic relation can also be combined for analysing more complex spatial effects, such as displacement, aggregation, general increase and general decrease. These four modules are designed based on a comprehensive literature review and two international-wide studies. The modules unite to form an integrated system, yet each can be operated independently. Case studies are presented in order to evaluate this system based on the collected real-world data from the Hong Kong police force and London's crime map. All the results are in line with expert predictions and corroborated by related empirical crime studies. In addition, three additional case studies were conducted from May to August 2015 in order to further evaluate the information generated from the four different modules in the Dynamic Pattern Analysis system (DPA system). Different methods, such as survey and clustering performance evaluation were used for different evaluation purposes. According to the evaluation results of the three additional case studies, we found that making use of DPA output (i.e. information generated from the DPA approach) was better than making use of other benchmark output (i.e. information generated from the corresponding benchmark approach) in different aspects. In case study 1, we invited fifteen international crime analysis professionals, including three highly-cited scholars in spatial-temporal crime analysis, to participate in an online survey. We found that making use of DPA output enhanced the accuracy of analysis. Moreover, the results of the survey also indicated that making use of DPA output not only changed international professionals' police resource allocation decisions significantly, but also the decisions made were more suited to the local situation. In case study 2, we used a two-year theft-related dataset to create three transformed datasets in three different forms of temporal pattern. For each transformed dataset we conducted clustering. The clustering results were evaluated by using the measures (i) purity and (ii) genetic distance. We found that using DPA's form of temporal pattern (i.e. cyclic temporal pattern) for clustering can create clusters with higher intra-cluster similarities and lower inter-cluster similarities than using other forms of temporal pattern. In case study 3, we invited fourteen experienced crime analysis professionals to participate in an online survey. The results of the survey indicated that making use of DPA output not only changed their police resource allocation decisions significantly, but also the decisions made were in line with the corresponding crime prevention guideline.