UCL logo

UCL Discovery

UCL home » Library Services » Electronic resources » UCL Discovery

Modelling of sparse spatio-temporal point process (STPP) - An application in predictive policing

Adepeju, MO; (2017) Modelling of sparse spatio-temporal point process (STPP) - An application in predictive policing. Doctoral thesis , UCL (University College London).

Full text not available from this repository.

Abstract

With a limited amount of resources, the new practice of predictive policing has exhibited great potential in relation to improving policing efficiency and reducing urban crime levels. The practice of predictive policing involves the use of analytical hotspot methods to identify the likely locations of future crimes, so that pre-emptive steps can be taken towards preventing those crimes from ever taking place. However, two critical issues have inhibited the prevalence of the practice amongst the law enforcement community. One, the lack of sufficient evaluation criteria for assessing and comparing the available methods so that the best selection can be made, and two, the inherent poor predictive performances of the available methods, as measured by some existing criteria. This thesis will contribute to the literature in this area by developing both a robust evaluation framework and an improved predictive framework to address the aforementioned issues. The analytical frameworks are designed based on the general modelling idea of the sparse spatio-temporal point process (STPP). The evaluation framework, which may be applied to any predictive hotspot method, comprises improved existing metrics and novel metrics. These metrics consider different performance aspects of predictive hotspots and are more comprehensive than what have been proposed in any previous studies. The predictive framework, referred to as ‘spatially adaptive time series hotspot (SATH)’, is specifically designed to demonstrate an improved performance over the existing methods, particularly in terms of the aspects of predictive accuracy and computational efficiency; two of the most critical factors that are key for prompt crime interventions. The utility of the two developed frameworks is demonstrated with two case studies of crime prediction carried out for the London Borough of Camden, London, UK, and the South-side Chicago, US. The first case study demonstrated that the developed evaluation framework is a robust assessment toolkit by which various performance aspects of different methods can be assessed and compared. The outcome of an assessment based on the framework can help a police department to determine the best method that is most suitable to achieve the performance objective of the department. The second case study demonstrated that the developed predictive framework (SATH) is better than the best-performing existing method, i.e. the self-exciting point process (SEPP), especially in terms of the predictive accuracy and computational speed. In terms of the accuracy, the SATH demonstrated statistically significant improvements (at p≤0.05) over SEPP in all data cases considered, while it is also more than 90 times faster, computationally. This research represents work of exceptional importance to the field of law enforcement; as it provides enhancements to the practice of predictive policing so that the safety and security of people can be ensured, in the face of increasingly limited resources.

Type: Thesis (Doctoral)
Title: Modelling of sparse spatio-temporal point process (STPP) - An application in predictive policing
Event: University College London
Language: English
Keywords: Spatio-temporal, Point process, Crime hotspot, Predictive Policing, Crime intervention
UCL classification: UCL > Provost and Vice Provost Offices > UCL BEAMS
UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science
URI: http://discovery.ucl.ac.uk/id/eprint/1555721
Downloads since deposit
2Downloads
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item