Abstract

Fire debris samples are currently analyzed according to ASTM E1618-19, which is the "Standard Test Method for Ignitable Liquid Residues in Extracts from Fire Debris Samples by Gas Chromatography-Mass Spectrometry." This method requires that an analyst make a visual comparison to an appropriate reference sample using the total ion and the extracted ion chromatograms. The analyst then provides an opinion about whether an ignitable liquid residue is present in the sample. The method is inherently subjective due to the visual interpretation that is needed. In order to automate this process, this work uses neural networks and a subset of the ions specified in ASTM E1618-19, which represent many of the compounds present in ignitable liquids, to cluster and classify ground-truth fire debris samples. The first part of this work demonstrates that these ions provide sufficient information to allow for the clustering of the ignitable liquid classes defined in ASTM E1618-19 and substrate pyrolysis extracts using self-organizing maps. Classification using self-organizing maps resulted in a 96% correct classification rate on an independent test set. The latter portion of this work demonstrates the use of the ASTM ions in conjunction with feedforward neural networks to evaluate laboratory prepared ground-truth fire debris samples. An optimal neural network model was selected from a set of candidate models that were trained on in-silico fire debris samples. Receiver operating characteristic curves were used to select an optimal decision threshold for classifying a fire debris sample as positive or negative for ignitable liquid residues using a false positive to false negative cost ratio of 10. The use of this threshold for classification resulted in a somewhat conservative model with a true positive rate of 0.59 and a false positive rate of 0.07 for a set of laboratory-generated ground-truth fire debris samples.

Notes

If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu.

Graduation Date

2021

Semester

Summer

Advisor

Sigman, Michael

Degree

Doctor of Philosophy (Ph.D.)

College

College of Sciences

Department

Chemistry

Degree Program

Chemistry

Format

application/pdf

Identifier

CFE0008745;DP0025476

URL

https://purls.library.ucf.edu/go/DP0025476

Language

English

Release Date

August 2021

Length of Campus-only Access

None

Access Status

Doctoral Dissertation (Open Access)

Included in

Chemistry Commons

Share

COinS