Combined target factor analysis and Bayesian soft-classification of interference-contaminated samples: forensic fire...

Area of Science:

• Forensic Science
• Analytical Chemistry
• Computational Statistics

Background:

• Fire debris analysis is crucial for arson investigation.
• Identifying ignitable liquid residues (ILRs) can be challenging due to complex sample matrices.
• Existing methods may struggle with background pyrolysis products.

Purpose of the Study:

• To develop and validate a novel Bayesian soft classification method for fire debris analysis.
• To enhance the accuracy of identifying specific classes of ignitable liquids (ASTM E1618).
• To assess the method's performance on diverse datasets, including laboratory and field burns.

Main Methods:

• Utilized Target Factor Analysis (TFA) on total ion spectra (TIS) from multiple fire debris samples.
• Employed a library of reference TIS from known ignitable liquids as target factors.
• Applied Bayesian decision theory with kernel-smoothed class-conditional distributions for classification.
• Incorporated soft classification to estimate probabilities of ILR presence, accounting for background noise.

Main Results:

• The Bayesian soft classification method demonstrated robust performance in analyzing fire debris.
• The method successfully identified ignitable liquid residues from specific ASTM E1618 classes.
• Achieved approximately 80% correct classification rates in both laboratory and large-scale field tests.

Conclusions:

• The developed Bayesian-TFA method offers a powerful tool for forensic fire debris analysis.
• The soft classification approach provides probabilistic assessments, improving confidence in identifying ILRs.
• This technique shows significant potential for improving the accuracy and reliability of fire investigations.