Bayesian network models
E.1 Introduction to Bayesian networks
There are plenty of LR models that could be used for measuring the evidential value of samples, for example in comparison and classification problems (Chapters 4 and 5). These models allow us to evaluate univariate and multivariate data and they also take into account various sources of errors, for example the variation between- and within-objects (Section 3.3.2). Moreover, these LR models allow us to apply a kernel density estimation procedure when the data are not normally distributed (Section 3.3.5). Such a situation is especially common for the between-object distribution in forensic databases. The disadvantage of applying these models is that they require a relatively large database in order to evaluate all the model parameters. This point is especially crucial for the evaluation of multivariate data. Moreover, there is little software available which can be used relatively easily by people without experience in programming to determine likelihood ratios (but see calcuLatoR; Appendix F). Therefore, if scientists want to use these models they have to write their own case-specific routines using the R software introduced in Appendix D, for example. Also, there is limited understanding of these evaluation procedures, such as those related to LR models, among the representatives of judicial systems, including forensic experts. Therefore, the possibility of applying models based on Bayesian networks (BNs) ...