An algorithm for forensic toolmark comparisons.

Area of Science:

• Forensic Science
• Materials Science
• Computer Science

Background:

• Traditional toolmark analysis relies on subjective human judgment, leading to inconsistencies.
• Numerous variables like angle and direction complicate accurate toolmark comparisons.
• Lack of transparency in forensic toolmark analysis can impact legal outcomes.

Purpose of the Study:

• To develop an objective, data-driven method for forensic toolmark analysis.
• To improve the reliability and transparency of toolmark comparisons.
• To create a standardized, open-source solution for forensic examiners.

Main Methods:

• Generated a dataset of 3D toolmarks from consecutively manufactured slotted screwdrivers at various angles and directions.
• Employed Partitioning Around Medoids (PAM) clustering to group toolmarks.
• Utilized Known Match and Known Non-Match densities to establish classification thresholds.
• Fitted Beta distributions to densities for deriving likelihood ratios for new toolmark pairs.

Main Results:

• PAM clustering demonstrated that toolmarks group by the manufacturing tool, not by the angle or direction of mark generation.
• Established classification thresholds with high accuracy.
• Achieved a cross-validated sensitivity of 98% and specificity of 96% in toolmark classification.

Conclusions:

• The developed approach offers a standardized and objective method for comparing toolmarks, specifically for slotted screwdrivers from similar manufacturing processes.
• This data-driven technique enhances the reliability of forensic toolmark analysis.
• The open-source solution has the potential to decrease miscarriages of justice by providing objective evidence.