Statistical discrimination of liquid gasoline samples from casework

Area of Science:

• Forensic Science
• Analytical Chemistry
• Chemometrics

Background:

• Gas chromatography-mass spectrometry (GC-MS) is crucial for analyzing complex mixtures like liquid gasoline.
• Differentiating liquid gasoline samples in forensic casework presents analytical challenges.
• Multivariate statistical methods offer powerful tools for pattern recognition in complex datasets.

Purpose of the Study:

• To develop and apply multivariate pattern recognition techniques for differentiating liquid gasoline samples from forensic casework.
• To assess the effectiveness of various supervised learning methods in classifying gasoline samples based on GC-MS data.

Main Methods:

• Utilized gas chromatography-mass spectrometry (GC-MS) for sample analysis.
• Employed supervised learning approaches including Principal Component Analysis (PCA), Canonical Variate Analysis (CVA), Orthogonal Canonical Variate Analysis (OCVA), and Linear Discriminant Analysis (LDA).
• Validated classification models using the jackknife cross-validation method to estimate error rates.

Main Results:

• Sufficient variability was found in the gasoline sample population to enable distinct differentiation.
• Canonical Variate Analysis (CVA) achieved complete sample differentiation using only three dimensions.
• Orthogonal Canonical Variate Analysis (OCVA) required four dimensions, while Principal Component Analysis (PCA) needed ten dimensions for accurate classification.

Conclusions:

• Multivariate pattern recognition effectively differentiates liquid gasoline samples in forensic contexts.
• CVA demonstrates high efficiency in classifying gasoline samples with minimal dimensionality.
• The developed procedures enhance the application of multivariate analysis in fire debris casework.