NCJ Number
240684
Date Published
October 2012
Length
45 pages
Annotation
The goal of this research was to develop a data-analysis method that can classify ignitable liquid residue in the presence of background interferences found in fire debris.
Abstract
The research developed a novel method for classifying ignitable liquid residues into the ASTM classes in the presence of a strong background signal. The method uses target factor analysis (TFA) in combination with Bayesian decision theory, which provides results in the form of posterior probabilities that a set of samples from a fire scene contain an ignitable liquid of a specific ASTM class; however, error rates are not currently available for fire debris analysis, other than extrapolations from proficiency tests. The method was further refined by introducing a sensitivity parameter that made the method conservative in its predictions. The method allows classification of a sample into multiple classes creating the possibility of not assigning the sample to any of the available classes. The work was divided into three tasks. Task I focuses on the development of the combined TFA and Bayesian decision theory, as well as the testing of the method by using computationally constructed data-sets prepared by using total ion spectra (TIS) from ignitable liquids and burned substrates from the Ignitable Liquids Reference Collection and Database and the Substrate Databases, respectively. Both are NCFS/TWGFEX databases. Task II tested the method by using laboratory-generated burn samples. Data-sets were produced by burning common building/furnishing materials with differing amounts of applied ignitable liquid, as well as by varying both the amount of applied liquid and the relative amount of substrate materials. Task II further tested the method on large-scale burn samples produced specifically for testing the methods. The 15-percent incorrect classifications included those samples in which the ignitable liquid residue was heavily weathered; and in some cases, the liquid may have completely evaporated. 15 figures, 6 tables, 24 references, 3 listings of publications disseminating the research, and appended supplementary data
Date Published: October 1, 2012