Title: Connecting actuarial judgment to probabilistic learning techniques with graph theory

Abstract: Graphical models have been widely used in applications ranging from medical expert systems to natural language processing. Their popularity partly arises since they are intuitive representations of complex inter-dependencies among variables with efficient algorithms for performing computationally intensive inference in high-dimensional models. It is argued that the formalism is very useful for applications in the modelling of non-life insurance claims data. It is also shown that actuarial models in current practice can be expressed graphically to exploit the advantages of the approach. More general models are proposed within the framework to demonstrate the potential use of graphical models for probabilistic learning with telematics and other dynamic actuarial data. The discussion also demonstrates throughout that the intuitive nature of the models allows the inclusion of qualitative knowledge or actuarial judgment in analyses.