ABSTRACT

This chapter gives an overview of the applications of Bayesian Networks (BNs) in forensic statistics. Also called Probabilistic Expert Systems, BNs constitute a valuable tool that for solving problems of forensic identification involving complex probabilistic argument and computation. Building a BN has two stages ⸺ constructing a graphical representation of the problem, involving a node for each relevant variable or hypothesis, with connections between these that encode the way in which they depend, probabilistically, on each other; and then adding quantitative information about those probabilistic dependencies. Once the BN is set up in a suitable software environment, one enters the available evidence on the observed variables, and the system computes the resulting remaining uncertainty about hypotheses of interest.

The chapter summarizes the basic logical structure of probabilistic argument in the legal context and introduces BNs for eyewitness evidence and forensic identification more generally. It describes an extension of BNs as OOBNs (object-oriented Bayesian networks), with applications to simple criminal identification, disputed paternity, and complex criminal cases involving family relationships (allowing for complications, such as the possibility of mutation, mixed DNA profiles, and representing uncertain allele frequencies and heterogeneous reference populations).