Compositional data analysis for elemental data in forensic science

Area of Science:

• Geochemistry
• Forensic Science
• Data Analysis

Background:

• Elemental composition analysis is crucial for material discrimination in forensic science.
• Traditional methods often overlook data constraints inherent in compositional datasets.
• Compositional Data Analysis (CoDa) offers a robust framework for handling such data.

Purpose of the Study:

• To apply CoDa analysis for discriminating materials based on elemental composition within a forensic context.
• To evaluate the effectiveness of CoDa in separating geological samples, specifically New Zealand nephrite.
• To develop an objective method for material interpretation in forensic investigations.

Main Methods:

• Utilized a compositional data (CoDa) analysis framework.
• Applied Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) for dimensionality reduction and classification.
• Reduced the full elemental composition to a descriptive subcomposition for enhanced discrimination.
• Employed a 10-repeat, three-fold cross-validation technique for model assessment.

Main Results:

• Achieved successful separation of in situ nephrite outcrops from a defined area.
• Demonstrated that a descriptive subcomposition, derived through CoDa, was more effective than the full composition for discrimination.
• The LDA classification model yielded a low mean error rate of 2.9% upon validation.
• The CoDa framework successfully addressed and managed the constraints of compositional data.

Conclusions:

• CoDa analysis provides an objective and effective framework for material discrimination in forensic science.
• The developed methodology offers a significant improvement over subjective pattern-matching approaches.
• This approach enhances the reliability and accuracy of forensic material analysis through rigorous data handling.