Grant Application: 

Advancing the use of the forensic-data-science paradigm for fired-cartridge-case comparison

Team members:

• Geoffrey Stewart Morrison, Aston University (Lead)
  
  
• Rachel Boton-King, Nottingham Trent University
  
  
• Flávio de Barros Vidal, University of Brasília
  
  
• Lehi Sudy dos Santos, Brazilian Federal Police
  
  

Summary

Across many branches of forensic science, current practice is based on human perception and subjective judgement, analysis and interpretation methods are non-transparent and susceptible to cognitive bias, interpretation is often logically flawed, and method validation is lacking. A new paradigm for evaluation of forensic evidence has been widely adopted for DNA, and has made substantial advances in forensic voice comparison. In the forensic-data-science paradigm, methods are based on relevant data, quantitative measurements, and statistical models / machine-learning algorithms. These methods are transparent and reproducible, are intrinsically resistant to cognitive bias, use the logically correct framework for interpretation of forensic evidence (the likelihood-ratio framework), and are empirically calibrated and validated under casework conditions. The likelihood-ratio framework is considered the logically correct framework for interpretation of evidence by the vast majority of experts in forensic inference and statistics, and by key organizations including the European Network of Forensic Science Institutes and the Forensic Science Regulator for England & Wales.

This research project aims to advance adoption of the forensic-data-science paradigm in fired-cartridge-case (FCC) comparison, an important task in firearms examination, a branch of forensic science in which the paradigm shift has so far made little progress.

When a firearm is fired, it leaves marks on the base of the cartridge case. There will be variability between the marks on different FCCs fired from different firearms (between-source variability), but there will also be variability between the marks on different FCCs fired from the same firearms (within-source variability). After a firearm is fired, the FCC is ejected, and it may later be recovered. A forensic practitioner may compare the marks on an FCC recovered from a crime scene with the marks on another FCC recovered from the same or a different crime scene, or with the marks on multiple FCCs fired from a known firearm (e.g., a firearm found in the possession of a suspect), and express a conclusion with respect to whether the FCCs were fired from the same firearm or from two different firearms. Current practice is based on visual inspection, using a comparison microscope, and subjective judgement. Currently, conclusions are expressed as “identification”, “inconclusive”, or “elimination”, which is inconsistent with the logic of the likelihood-ratio framework.

In this research project:

1.    We will collect a dataset of high-resolution 3D images of FCC bases representative of within-source and between-source variability in marks left by firearms belonging to a class of firearms commonly encountered in casework. Using a 3D imaging system commonly employed in forensic laboratories, we will image 17,000 FCCs, 10–14 FCCs fired from each of 1,300 firearms. This will allow us to model between-source and within-source variability.

2.    We will develop and validate a forensic-FCC-comparison system suitable for evidential casework use. The system will be based on state-of-the-art machine-learning methods and state-of-the-art forensic-inference methods, including adapting methods used in forensic voice comparison. The system will output well-calibrated likelihood ratios as quantifications of strength of evidence. The likelihood ratios will quantify the probability of obtaining the marks on the FCCs if the FCCs were fired from the same firearm versus the probability of obtaining the marks on the FCCs if the FCCs were fired from two different firearms belonging to the same class of firearms.

3.    We will compare the performance of the statistical-model/machine-learning-based system with current human-based methods.

4.    We will conduct end-user testing.

Status:

Application submitted 2024-05