Validation of Bayesian networks- with a case study on fingerprint general patterns

Abstract

In the first part of the project we design and validate Bayesian networks for evaluation of the fingerprint general pattern. These networks have two applications in forensic science. Firstly, it will help the fingerprint examiner reduce the number of reference fingerprints he has to search (to find the donor of the fingermark), by ascertaining from which finger(s) is it more likely for the general pattern(s) of the fingermark(s) to occur. Secondly, it will help the fingerprint examiner to quantify the strength of evidence of the general pattern, in terms of likelihood ratios.

To perform validation, we sometimes need to find the best explanation for a set of evidence. In Bayesian network, finding the best explanation amounts to finding a value assignment to some of the variables in the network that has highest posterior probability given the available evidence (i.e. the best explanation is a most likely one). This problem is known as most probable explanation or maximum a posterior assignment which is NP-hard in general. In the second part of this project, we propose two heuristics and based on experiments on some synthetic data, we show that they converge approximately to the true MAP assignment.