The assessment of different items of evidence is a challenging process in forensic science, particularly when the relevant elements support different inferential directions. In this study, a model is developed to assess the joint probative value of three different analyses related to some biological material retrieved on an object of interest in a criminal case. The study shows the ability of probabilistic graphical models, say Bayesian networks, to deal with complex situations, those that one expects to face in real cases. The results obtained by the model show the importance of a conflict measure as an indication of inconsistencies in the model itself. A contamination event alleged by the defense is also introduced in the model to explain and solve the conflict. The study aims to give an insight in the application of a probabilistic model to real criminal cases.

Analysis of marks recovered from different crime scenes can be useful to detect a linkage between criminal cases, even though a putative source for the recovered traces is not available. This particular circumstance is often encountered in the early stage of investigations and thus, the evaluation of evidence association may provide useful information for the investigators. This association is evaluated here from a probabilistic point of view: a likelihood ratio based approach is suggested in order to quantify the strength of the evidence of trace association in the light of two mutually exclusive propositions, namely that the n traces come from a common source or from an unspecified number of sources. To deal with this kind of problem, probabilistic graphical models are used, in form of Bayesian networks and object-oriented Bayesian networks, allowing users to intuitively handle with uncertainty related to the inferential problem.