Evaluation and comparison of methods for forensic glass source conclusions.

Area of Science:

• Forensic Science
• Materials Science
• Analytical Chemistry

Background:

• Float glass is crucial forensic evidence in criminal investigations.
• Chemical composition analysis, often via laser ablation inductively coupled mass spectrometry (LA-ICP-MS), distinguishes glass fragments.
• Machine learning offers a novel approach to probabilistic source conclusions for forensic glass analysis.

Purpose of the Study:

• To evaluate machine learning classifiers for forensic float glass analysis.
• To understand the performance of learning algorithms compared to traditional methods.
• To identify limitations in current ASTM standards for glass comparison.

Main Methods:

• Simulated forensic scenarios using a database of glass elemental concentrations.
• Application and analysis of two distinct machine learning classifiers.
• Examination of the ASTM International standard decision process for glass comparison.

Main Results:

• Machine learning methods demonstrated lower classification error rates than traditional approaches.
• Analysis revealed that the standard ASTM method may not be optimal for source conclusions.
• The study highlights the potential of advanced algorithms in forensic glass profiling.

Conclusions:

• Machine learning algorithms show promise for improving the accuracy of forensic glass analysis.
• The current ASTM standard for comparing chemical compositions needs re-evaluation.
• More data is essential for developing robust comparison rules for float glass based on elemental concentrations.