Graphical Models for Forensic Analysis

Here we are concerned with systems to assist in the evaluation of ev- idence presented in a criminal or civil court case. Such a case may have a mixed mass of evidence of many kinds, all of it subject to un- certainty. We describe how such a case can be helpfully represented by means of a Bayesian Network (BN), or Probabilistic Expert System: a directed graphical model describing the various items of evidence and hypotheses, and the probabilistic relationships between them. Such a representation displays clearly the relevance of the evidence to ques- tions of interest, and supports ecient routines to compute the impact of the evidence presented. In many cases the BN can be constructed as an object-oriented Bayesian network (OOBN), a top-down hierarchical structure which hides irrelevant detail and simpli es both construction and interpretation.