Predictive analytics and Policing: Translating cutting-edge academic research into actionable intelligence and developing useable software tools

Lead Research Organisation: University College London
Department Name: Security and Crime Science

Abstract

Summary

Criminological research has for some time focused on the types of people that commit crime and why they might do so. However, over the last decade, there has been substantial progress in research concerned with the complimentary question of whether there exist regularities in how, where and when crimes are committed. Such research has demonstrated that the risk of crime is not random, but that the risk of crime is greater for some people, in particular neighbourhoods, and at some homes. Similarly, research demonstrates that the risk of crime varies over time. For instance, crime is more likely at certain times of the day and months of the year. These findings have clear implications for operational policing and currently inform the crime reduction strategies adopted by many UK police forces.

More recent academic research conducted by the applicants demonstrates that other regularities in crime patterns exist and that their identification can inform police decision-making in such a way so as to lead to the more effective use of police (and other) resources. For example, our research shows that when a burglary occurs at one location, others are more likely to occur at the same location or nearby in the near future. Put another way, the risk of burglary spreads in much the same ways as a communicable disease. Importantly, the identification of such regularities can provide intelligence that can inform what the police might do to prevent or detect crime. Furthermore, based on such findings, it is possible to develop mathematical models to generate predictions regarding the future locations of crime events. Such models have been shown to outperform existing methods of crime mapping.

The ability to identify these kinds of regularities and to use them to inform police priorities or tactics is particularly important at a time when police budgets are being cut. However, because the types of analysis noted above are at the cutting edge of academic research, and were generated for research purposes, they cannot be conducted using available software. The aim of the proposed project is to bridge this knowledge-practice gap, identifying the most promising types of analysis and developing usable software packages that would make a real difference to police practice.

The collaboration would facilitate a two-way exchange between police practitioners and academics and lead to the translation of cutting edge research into tools that can be used to prevent harm to society. We envisage that the project would also lead to the identification of new research questions that would inform the academic agenda.

Planned Impact

The results of this work would benefit police practitioners and academics alike, by facilitating a two-way knowledge exchange as part of an effort to reduce crime. The undertaking of such a project will itself help to develop and strengthen the dialogue between researchers and the police. Ultimately, results from this project should benefit society at large, by taking leading edge research and translating the methods into tools that could be used to prevent harm to individuals, households and neighbourhoods. Indeed, the main aim of the proposed project is to bridge a knowledge-practice gap and make a real difference to police practice.

To recapitulate, we intend to take some of our more leading edge methodological approaches and develop these into tools that the police can actually use on a daily basis. However, developing such tools is only the first step in what will hopefully be a longer-term engagement with police practitioners about how academic research and analysis can inform police strategy and tactics. The proposed project will pave the way for this form of collaboration and will demonstrate how such interaction can help develop the crime reduction research base in innovative ways.

Like many sectors, the Police are currently struggling with budget cuts and a consequent squeeze on available resources. Identifying ways to inform police priorities or tactics, including the most efficient allocation of deployments, is crucial to their continuing performance. Similarly, researchers are being asked more about the relevance of their work in the every day. The opportunity to work with practitioners and to directly elicit and discuss feedback and direction for what are the most pressing and relevant concerns for practitioners, will lead to research which is better attuned to those current problems that the police face. Thus, we envisage that the project would lead to the identification of new research questions that would inform the academic agenda.

The research outputs will be developed in close conjunction with our project partners, and will be tailored to their needs. However, as analysts around the country work with similar data on similar data bases, we expect that the software solutions developed will be (to some extent) transferable across police forces, albeit with some additional effort. In addition, there is a wider community of analysts working in the UK today that look at crime data as part of their day to day work; these are analysts that work for community safety partnerships, the fire service, emergency services and transport organisations (amongst others). They use similar methods and software for analysing problems in their respective domains. These analysts also work with similar data to the police (often the same data) and are concerned with many overlapping themes. This type of work could serve as a model for future collaborations of analysts with academia and the products from this work could be equally applicable in the analysis work done in these related domains.

Similarly, we plan to write a series of short one to two page notes that summarise the cutting edge research for the analysts and officers to save them reading the research papers. These papers could be developed as a resource for the wider police and analyst community, by making them availabe on our website for general consumption.

At the end of the project, we envisage applying for follow-on funding to generate programming code for further tools and to generate more generic versions of those tools already developed. These more generic versions could be made freely available to other police forces and there is the further potential to commercialize these tools for the private sector. The lessons learned as part of the initial project will inform and shape those agendas.

Publications

10 25 50

publication icon
Bowers K (2014) The Handbook of Security

publication icon
Johnson S (2014) How do offenders choose where to offend? Perspectives from animal foraging in Legal and Criminological Psychology

publication icon
Johnson, S.D. (2014) Space-Time Pattern Analysis

publication icon
Johnson, S.D. (2015) Predictive policing

publication icon
Johnson, S.D. (2013) Predictive Policing in 3rd International Policing Conference Scotland (Edinburgh)

publication icon
Johnson, S.D. (2014) Crime pattern theory and the influence of street networks on burglary risk (Invited Talk) in 1st International Summit on Scientific Criminal Analysis, SANTIAGO, CHILE

publication icon
Johnson, S.D. (2014) The Space-Time Dynamics of Crime (Invited Talk) in 1st International Summit on Scientific Criminal Analysis, SANTIAGO, CHILE

publication icon
Johnson, S.D. (2014) Street Networks and Risk

publication icon
Rosser G (2017) Predictive Crime Mapping: Arbitrary Grids or Street Networks? in Journal of quantitative criminology

 
Description A sample of police analysts and officers were interviewed about how analysis is perceived and used in operational policing, if they thought predictive analytic software would be useful, and what this might do. There was perceived to be cultural acceptance of analysis as integral to tactical and strategic police decision-making. Beyond crime mapping, predictive analytics were not used. Apropos the utility of crime analysis, there was a perception amongst some that officers do not always commission the right analytic products. Some thought that officers were not always aware of the types of analysis possible, or requested in-depth analyses for patterns that might reflect random fluctuation. Officers expressed a desire for more evidence of what works to reduce crime in analytic products generated. The discussions suggest that as well as incorporating training on what works to reduce crime, police professional development should consider what forms of analysis are possible, when they are appropriate (and in what depth), and how these can inform crime reduction.

When asked about the automation of tasks, analysts noted they are sometimes required to testify in court, and that they would be more confident about analyses completed manually. Potential evidential issues such as these are often overlooked in academic research but should inform how practical applications - including software - that result from it are developed.

Interviewees were asked about the potential utility of, and their suggestions about, four analytic approaches: 1) mapping crime at the street (rather than area) level to identify "hot streets"; 2) quantifying how crime risk spreads across different police areas and across offence types; 3) identifying crimes that - due to their proximity in space and time - may be "linked" to an offender; and, 4) predicting on which streets crime is next most likely.

There was consensus that, compared to traditional area-level mapping, the "hot streets" tool would be very useful, particularly if it incorporated contextual data from other organisations. One suggestion was that streets (rather than crime types) could be the focus of crime analysis, with crime prevention addressing "problem" roads through collaboration with other organisations.

Some interest was expressed in the second two approaches, particularly for the analysis of large volumes of data. Surprisingly, the initial reaction to the predictive mapping software was mixed. Some felt that predictions produced for longer timescales (1-3 months) would more usefully inform police deployments, whilst others felt that short-term predictions would be very helpful.

Based on the feedback, a software toolbox was developed. Analyses can be conducted for particular (selectable) policing areas, types of offences, and intervals of time (to show patterns of change). Maps produced can be customized and patterns shown against an OpenStreetMap of the area (automatically imported). Each of the four analytic methods were implemented - all novel. Feedback on the toolkit was very positive and further suggestions were made to develop it. Due to force IT issues, the software was not tested operationally, but exceeding the project objectives, elements of the toolbox have been made freely available.
Exploitation Route Police forces, community safety partnerships and those involved in crime prevention and security more generally should consider conducting crime pattern analysis at the street segment rather than area level. Software developers should consider developing analytic tools that facilitate such analysis. Some of the software developed for this project has been made freely available, police analysts might use that software to analyse crime patterns in their local area.
Sectors Communities and Social Services/Policy

Digital/Communication/Information Technologies (including Software)

Government

Democracy and Justice

Security and Diplomacy

Other

 
Description The core aims of this knowledge exchange award were to examine the use of predictive analytics in policing, to explore what types of approaches police officers and analysts perceived might be the most useful, to develop software that might meet this perceived need, and to assess the perceived utility of the software. Most well received was software that would generate predictions of the streets on which crime is most likely to occur over the next few days. Since the ESRC project concluded, this approach has been embraced by a police force with whom we worked to develop the software for use in an operational context. The software was initially tested in a few areas but was subsequently rolled out and made available to over 1000 police officers across the police force area. The work, which is a direct result of the knowledge exchange project, will contribute to the evidence base regarding the effectiveness of predictive policing approaches and police deployment strategies, thereby informing knowledge about "What works" to reduce crime.
Sector Other
Impact Types Policy & public services

 
Description Multiple citations within Houses of Parliament, Parliamentary Office of Science & Technology POSTNOTE (470, July 2014). The work cited was the focus of the ESRC knowledge exchange grant ES/K000721/1.
Geographic Reach National 
Policy Influence Type Citation in other policy documents
URL http://www.parliament.uk/business/publications/research/briefing-papers/POST-PN-470/big-data-crime-a...
 
Description Crime reduction
Amount £29,000 (GBP)
Organisation Police and Crime Commissioner for West Yorkshire 
Sector Public
Country United Kingdom
Start 08/2014 
End 09/2016
 
Title UCL Near Repeat Toolkit 
Description The software is used to analyse spatial crime data. It is an implementation of the univariate Knox test (Knox, 1964), and the Bivariate Knox test (Johnson et al., 2009). It also has simple (but novel) mapping and querying functionality. Principally, it allows the user to analyse crime (or other event data) to detect space-time clustering for a particular type of crime, or (in the bivariate case) across two crime types. That is, it allows the use to test whether a sample of crimes are more likely to occur near to and shortly after a previous event than would be expected, if the timing and location of crimes were unrelated. The data is also mapped against an Open Street Map background. Where crimes occur near to and after another, an arrow between the two indicates which came first. Knox, E. G., & Bartlett, M. S. (1964). The detection of space-time interactions. Journal of the Royal Statistical Society. Series C (Applied Statistics), 13(1), 25-30. Johnson, S. D., Summers, L., & Pease, K. (2009). Offender as forager? A direct test of the boost account of victimization. Journal of Quantitative Criminology, 25(2), 181-200. 
Type Of Technology Software 
Year Produced 2015 
Impact After considerable development, the tool was published March 2016. Consequently, at the time of reporting, it is too soon for impacts to have been realised. 
URL http://www.ucl.ac.uk/jdibrief/analysis/space-time-pattern-analysis
 
Description Features interview in the New Scientist 
Form Of Engagement Activity A magazine, newsletter or online publication
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Public/other audiences
Results and Impact Features interview in the New Scientist (13 Jul 2019) in which I discussed the security of the Internet of Things and other emerging crime and security risks that may lead to "crime harvests".
Year(s) Of Engagement Activity 2019
URL https://www.sciencedirect.com/science/article/pii/S0262407919312874
 
Description Presentation at Westminster Briefing organised event in London 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Professional Practitioners
Results and Impact Approximately 50 police officers of all ranks attended a Westminster Briefing organised event (see, http://www.westminster-briefing.com/fileadmin/westminster-briefing/Agendas/Predictive_policing.pdf) in London entitled"Predictive Policing: Reducing Crime Using Data and Technology". I gave a talk at this event, fielded questions, and engaged in debate with delegates regarding the use of crime forecasting in operational policing. This was the second event on this topic organised by Westminster Briefings at which I have spoken.

The event stimulated debate on the utility of predictive policing in an operational context and the exchange of ideas between officers from different police forces.
Year(s) Of Engagement Activity 2014
URL http://www.westminster-briefing.com/fileadmin/westminster-briefing/Agendas/Predictive_policing.pdf
 
Description Presentation on Predictive Policing for a Westminster Briefing 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Professional Practitioners
Results and Impact This was the first of two events organised by Westminster briefings at which I spoke. This event, entitled "Predictive Policing: Reducing Crime Using Data and Technology" took place in February 2014. Approximately 50 police officers of all ranks attended the event. I gave a talk at this event, fielded questions, and engaged in debate with delegates regarding the use of crime forecasting in operational policing.

The event stimulated debate on the utility of predictive policing in an operational context and the exchange of ideas between officers from different police forces.
Year(s) Of Engagement Activity 2014
 
Description Presentation on Predictive policing to the National Crime Analyst Working Group 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Professional Practitioners
Results and Impact My talk focused on the development of, and application of methods of predictive policing. Participants discussed the approach and shared experiences of crime reduction approaches informed by such methods.

After my talk, I was contacted by the police and other agencies about the approach and also about evaluation research. I was invited to give a further talk at a meeting in December 2015. Unfortunately, I cannot disclose the details of that meeting.
Year(s) Of Engagement Activity 2015