Bayesian models of criminal activities
Jim Smith (University of Warwick)
Friday 26th October, 2018 15:00-16:00 Maths 311B
Over recent years both in forensic science and criminology there has been an immense interest in supporting investigations using justifiable and verifiable probabilistic methods. One challenge though is that the domain experts in these fields are often not trained to think statistically. So the faithful representation of their knowledge and reasoning in a way that can be harmonised with an empirical data analysis is a challenge. One of the most successful ways of achieving such a translation is to use the halfway house of a graph which can evocatively express the processes within the domain whilst also providing a sound formal framework for this translation into a statistical model. In this talk I will describe how a recent class of graphical models based on event trees, the CEG, can be used to help translate hypotheses about how criminals are motivated and act into a probability model that can then be used as the basis within which to weigh evidence. The talk will be illustrated by three examples - hypotheses associated with a rape case, activities associated with electronic grooming and radicalisation leading to violence. We show how expert judgments and empirical evidence can be drawn together using this tool.