Download or read book Statistical Assessment of the Probability of Correct Identification of Ignitable Liquids in Fire Debris Analysis written by United States. Department of Justice and published by . This book was released on 2015 with total page 114 pages. Available in PDF, EPUB and Kindle. Book excerpt: Identification of ignitable liquid residues in the presence of background interferences, especially those arising from pyrolysis processes, is a major challenge for the fire debris analyst. The proposed research will lead to a mathematical model that allows for the detection of an ignitable liquid in a fire debris sample and the classification of the ignitable liquid according to the ASTM E1618 classification scheme. The research will examine the influence of substrate pyrolysis and non-pyrolysis interferences on: (1) probability of correct prediction of the presence of an ignitable liquid in real and simulated fire debris samples (Type I and Type II error rates) and (2) probability of correct prediction of the associated ignitable liquid ASTM class and sub-class (heavy, medium or light) in positive samples. Potential alternative sub-groupings of ignitable liquids will be examined based on cluster analysis techniques. Models will be examined which are based on principal components analysis (PCA), linear discriminant analysis (LDA) and soft independent model classification analogy (SIMCA). The model will be developed from the summed ion spectra of nearly 500 ignitable liquid and 50 pyrolysis sample GC-MS data sets with ANOVA-assisted variable selection. Training data sets will be taken from the National Center for Forensic Science ignitable liquid and substrate pyrolysis databases. Simulated fire debris samples generated in the laboratory and samples from large-scale burns will also be employed in model testing. Model performance will be statistically evaluated by receiver operator characteristic analysis. The final model will be implemented in a software solution for forensic laboratory use. This project proposed to investigate the development of a method for classifying fire debris GC-MS data sets as: (1) containing or not containing an ignitable liquid, (2) classifying any ignitable liquid that may be present under the ASTM E1618 classification scheme and (3) estimating the statistical certainty of the answers to questions 1 and 2. The proposed approach is to build a mathematical model that can correctly classify GC-MS data from ignitable liquids and pyrolyzed substrates (wood, plastic, etc.). The model will then be applied to GC-MS data from laboratory-generated fire debris samples, as well as ignitable liquids and substrates that were not used to build the model. The classification success of the model will allow a determination of the statistical performance of the model by ROC analysis. The model will be developed based on the total ion spectrum, which has already shown a propensity for classifying a set of ignitable liquids drawn from multiple ASTM classes.