Data integration for tracking criminal threats
In this project researchers are designing a decision support system that uses crime data to inform policing of serious, typically collaborative crime.
Organised criminal activities are becoming increasingly sophisticated, international and complex and are informed by a plethora of diverse and often very noisy data.
Through collaborations built up within the Turing, Professor Jim Smith is leading a project to design a user friendly decision support system. This system will integrate data to inform policing of serious, typically collaborative crime. The support system will use a probabilistic model to merge information in real time within three streams.
These streams are made up of the histories of suspects criminal activity, their networking activities and finally geographical and demographical information concerning the whereabouts of both threatening groups and also vulnerable infrastructure and victims. Each of these three components will be represented back to the client by different evocative graphs that will continually update in real time as threats develop.
The probabilistic methods ensure that all relevant information is used by each of the three component models and that assessments concerning each are consistent with one another. In this way information can be merged to provide more effective and further integrated police protection.
One of the big challenges in statistics and machine learning is how to model large complex systems in a feasible but justified way when these systems are informed by sometimes huge data sets. One of the best techniques for addressing these multifaceted problems is to model these with formally defined graphs. There graphs – collections of identified nodes connected by edges - can form a framework for making inferences efficiently and in real time when addressing dynamically changing environments.
Throughout Prof. Smith's career, he's designed many such bespoke graphical frameworks for different systems. Examples include models which integrate information associated with an accidental release of nuclear contamination after an accident at a plant, predict world trading of oil after a war, probability models associated with complex criminal cases with extensive supporting forensic evidence of different kinds, public health monitoring especially with regard to children’s health and most recently models to predict and control food poverty within the UK.
Prof. Smith designs these bespoke probabilistic graphical models to capture the dependence structures implicit in such processes. These can then be used to hang our more quantitative data based judgements to come to evidence based decision support for various players.
Dr Jim Q. Smith, Professor of Statistics, University of Warwick