Differentiation and evaluation of evidence value of styrene acrylic urethane topcoat car paints analysed by pyrolysis-gas chromatography

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

“…A proper evaluation process, e.g. by applying the LR models in References [2][3][4][5][6][7][8], requires that all factors important for evaluation, including information about the rarity of features and the correlation between them, be reliably estimated from available data. If there is not enough data then the forensic expert could evaluate data in a very subjective way by visual analysis of the available data supported by his knowledge.…”

“…Pyrolysis GC/MS (Py-GC/MS) [14,16]. Also, the chromatographic data used in the research were analysed by applying a likelihood ratio model and relatively low numbers of false positive and false negative answers were observed in Reference [8]. Once again, the results of such an analysis are burdened by a relatively large uncertainty since parameter estimation is based on 36 samples described by seven variables.…”

“…Data from 35 analysed samples (50-100 mg) of styrene acrylic urethane topcoat paints published in Reference [8] were used in the presented research. The samples were pyrolysed using CDS Pyroprobe 2000 pyrolyser (Analytix Ltd, UK) at 7508C for a duration of 20 s and then chromatographic analysis was performed (Auto System XL, Perkin Elmer, USA).…”

“…how often a certain rank was observed within the particular variable in a database of analysed car paints. Therefore, this node also allows for the inclusion of information on the rarity of a particular state, which should be included in the evaluation of the evidential value in a comparison problem [1,[3][4][5]8,9]. The node E contains conditional probabilities that allow the expression that a certain combination of ranks of discrete variables in the recovered (E ¼ Vr) and control (B ¼ Vc) samples could occur under each of the considered propositions:…”

Hybrid approach combining chemometrics and likelihood ratio framework for reporting the evidential value of spectra

“…In the presented approach the multidimensional problems were split into multiple problems of 90 lower dimensions, for which LR models are more credible. However, the applied methodology is only successful for the physicochemical data sets described by only a few variables [3,11]. Even though application of the graphical models is reported also for highly multivariate data [12,13], the LR models based on their outcomes may not be reliable enough for legal processing.…”