Progress toward the determination of correct classification rates in fire debris analysis

Area of Science:

• Forensic Science
• Analytical Chemistry
• Pattern Recognition

Background:

• Accurate identification of ignitable liquid residue (ILR) is crucial in fire investigation.
• Existing methods for ILR classification can be complex and time-consuming.
• Standardized classification systems, such as ASTM E 1618-10, are essential for reproducible results.

Purpose of the Study:

• To develop and validate a multistep classification procedure for identifying ignitable liquid residue in fire debris.
• To assign detected ignitable liquid residue into classes defined by ASTM E 1618-10.
• To assess the performance of the classification procedure using both simulated and real-world fire debris samples.

Main Methods:

• Utilized principal components analysis (PCA), linear discriminant analysis (LDA), and quadratic discriminant analysis (QDA).
• Developed a multistep classification procedure based on time-averaged mass spectra (total ion spectra).
• Tested the procedure using model datasets of commercial ignitable liquids, substrate pyrolysis products, and actual fire debris samples from laboratory and field burns.

Main Results:

• The optimal classification model achieved a true-positive rate of 81.3% for cross-validation samples.
• For actual fire debris samples, the true-positive rate was 70.9%.
• The false-positive rates were 9.9% for cross-validation and 8.9% for fire debris samples.

Conclusions:

• The developed multistep classification procedure demonstrates a viable approach for identifying and classifying ignitable liquid residue in fire debris.
• The method shows promising accuracy, although further refinement may improve performance on real-world samples.
• This approach supports the standardized classification of ignitable liquids as per ASTM E 1618-10, aiding fire investigation.