Evaluation of the evidential value of physicochemical data by a Bayesian network approach

Abstract

The growing interest in applications of Bayesian networks (BNs) in forensic science raises the question of whether BN could be used in forensic practice for the evaluation of results from physicochemical analysis of a limited number of observations from flammable liquids (weathered kerosene and diesel fuel) by automated thermal desorption gas chromatography mass spectrometry (ATD-GC/MS), car paints by pyrolysis gas chromatography mass spectrometry (Py-GC/MS) and fibres by microspectrophotometry (MSP) in the visible (VIS) range. Therefore, various simple BN models, which allow the evaluation of both discrete and continuous types of data, were studied in order to address questions raised by the representatives of the administration of justice, concerning the identification and classification of objects into certain categories and/or the association between two items. The results of the evaluation performed by BN models were expressed in the form of a likelihood ratio, which is a well-documented measure of evidential value in the forensic field. From the results obtained, it can be concluded that BN models seem to be promising tool for evaluating physicochemical data.

Likelihood Ratio Models for Classification Problems

“…Research on the usefulness of this model is described in Zadora (2010). The aim was to solve a classification problem, namely, whether the sample falls within the category of kerosene or diesel fuel.…”

Evaluation of Forensic Data Using Logistic Regression-Based Classification Methods and an R Shiny Implementation

“…A fundamental task for the forensic experts is that the results of the analyses, which have been performed on the collected pieces of evidence, have to be expressed in a very clear and straightforward way that can be easily shown in courtrooms and that can be immediately, where possible, understood even by non-specialists. However, at the same time, the applied statistical methodologies for data evaluation have to be rigorous and should not compromise the role that the forensic expert plays in the administration of justice (Zadora, 2010).…”

“…Another relevant feature of applying the LR in the field of forensics is that it overcomes the so-called "falling off a cliff " problem related to the traditional approach of using cut-off values in classification models (Gill et al, 2006;Pragst et al, 2010;Zadora, 2010;Robertson et al, 2016). In particular, the use of LR avoids the necessity of defining thresholds (such as the largely adopted p-value = 0.05 for a significance level of 95%).…”

Geochemical wolframite fingerprinting – the likelihood ratio approach for laser ablation ICP-MS data

“…This approach has also been used in the analysis of earprints, fingerprints, firearms, and tool marks, hair, documents, and handwriting (review can be found in [ 9 ]), as well as speaker recognition [ 21 ]. An increasing number of applications of this approach is found in the evaluation of physicochemical data recorded for microtraces of glass [ 12 – 14 , 22 – 27 ], explosives [ 28 ], car paints [ 29 – 33 ], polymers [ 31 , 32 ], fire debris [ 34 ], inks [ 35 , 36 ], fibers [ 29 ], drugs [ 37 – 39 ], food samples [ 40 , 41 ] and biological samples [ 42 ].…”