Bayesian Networks for Probabilistic Inference and Decision Analysis in Forensic Science, 2nd Edition

3.5 Evaluation using graphical models

3.5.1 Introduction

As noted in Chapter 1, Bayesian networks have been found useful in assisting human reasoning in a variety of disciplines in which uncertainty plays an important role. In forensic science too, Bayesian networks have been proposed as a method of formal reasoning that could assist forensic scientists to understand the dependencies that may exist between different aspects of forensic findings. Historically, Aitken and Gammerman (1989) were amongst the first to suggest the use of graphical probabilistic models for the assessment of scenarios involving scientific findings. Ideas expressed there have been developed by authors such as Dawid and Evett (1997) or Garbolino and Taroni (2002). These studies provide clear examples of the relevance of Bayesian networks for assisting the evaluation of forensic findings, but explanations, if any, as to how practitioners should use the method to build their own models, are mostly brief and very general. The problem is well posed in Dawid et al. (2002), where the authors note that finding an appropriate representation of a case under examination is crucial for several reasons (viability, computational routines, etc.) and that the graphical construction is to some extent an art form, but one which can be guided by scientific and logical considerations. Whilst no explicit guidelines exist as to how one should proceed in the construction of a Bayesian network, there are some general concepts ...