Organised criminal activities are becoming increasingly sophisticated, international, and complex. Policing such criminal threats is increasing informed by a range of diverse but often very noisy and biased data sources that now complement more traditional forms of intelligence. This project uses expert judgments to build frameworks that can systematically integrate data into bespoke decision support systems. These systems will aid police in effective resource allocation and in real-time tracking and combatting of a variety of criminal threats.
Explaining the science
Information concerning assault crime can be broadly categorised into three distinct types:
- Information about the histories and continuing threats posed by both specific identified individuals and the populations from which they are drawn, including individual criminal histories.
- Real time networking activities, such as social networking and web usage, of such threatening people.
- Real time geographical information about previous incidents, the location of threatening gangs, and the current protection of vulnerable infrastructure and victims.
Although each of these data types are linked and inform each other, relevant data of the different types is typically recorded and maintained under separate platforms. One of the best techniques for addressing the problem of modelling large complex systems informed by data streams like these is to use ‘graphs’ – a symbolic representation of a network and of its connectivity, simplified as a set of linked nodes.
Each of the three different platforms discussed above can be expressed through its own type of graph, each chosen to capture the specific characteristics of its domain. The bespoke graph can then provide a framework for efficient, real-time processing of information about criminal threats. This is particularly useful when addressing the dynamically changing environments encountered when policing criminal threats.
Through collaborations built up within the Turing two main strands of work have developed.
The first strand, working with stakeholders from government and the West Midland Police, involves scoping the requirements for a decision support system these stakeholders might employ. This includes ascertaining the nature and recording method of data and intelligence that might be used in the course of an investigation. From this, bespoke graphical models for each type of information are being designed.
The second strand is related to the requirements for algorithms that integrate decision support systems of various types into a coherent framework for probability modelling. These algorithms require an overarching system to be built that transfers information from one decision system to another. In this strand, technology is being developed to provide decision support for assault crime prediction and mitigation in the pursuit phase. The support system will use a probabilistic model to merge information in real time from the three major data streams described in ‘Explaining the science’.
Information from these streams will be presented to police in meaningful graphs that will continually update in real time as threats develop. The probabilistic methods ensure that all relevant information from each data stream is utilised and that assessments concerning each stream are consistent with one another. In this way information can be merged to provide more effective and integrated policing.
The researchers are currently in active collaboration or in discussion with various policing organisations. The support systems being developed have the potential to have worldwide impact in guiding more effective usage of resources and better co-ordinated information systems.
As of May 2018, the first of three graphical frameworks from the first strand of work has been customised to model violent crime, providing a new streamlined representational framework. The first paper to report this work will appear at the end of May 2018, targeted at a potential user.
Another paper developing this new technology for a machine learning audience will be reported later in 2018. Details of the other two components of the composite system are now under development.
The research has also recently shown that graphical frameworks enable information about the rationality of an adversary – here the criminal – to be embedded into the structure of the system for better calibration.
For more information, please contact The Alan Turing Institute