Thesis

Comparative forensic examination of fibres and gunshot-residue by vibrational spectroscopy and machine learning

Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2024
Thesis identifier
  • T17060
Person Identifier (Local)
  • 201851736
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Forensic science is the application of scientific disciplines such as chemistry in law enforcement to achieve justice. Trace evidence is minuscule physical materials found at crime scenes on a microscopic level. The role of classifying trace evidence is to collect the evidence from crime scenes and compare it to the relevant belongings of suspects. Fibres and gunshot-residue show significant value as trace evidence found in crime scenes. The most common fibre types involved in different crimes worldwide, as per published cases, are cotton, polycotton, denim, viscose, polyester, and polypropylene. These fabrics are the focus of this research. The textile fabrics were examined by two handheld near-infrared (NIR) spectrometers, SCIO® and NIRscan Nano, for in situ comparison of fibres, demonstrating capability at a crime scene. SCIO and NIRscan Nano data were preprocessed and modelled using PRFFECTv2 software, which showed good predictive accuracy. The accuracy, sensitivity, and specificity were in the range of 78-100% for the best binary classification models (one class versus others) and within the range of 65-100% for the best multi-class classification models. Also, the same methodology and fabrics materials were used for analysis in the presence of common contaminants (blood, rainwater, seawater, sand and gunshot-residue) to evaluate the performance of two NIR spectrometers for in situ analysis of different crime scene conditions. The models showed accuracy, sensitivity, and specificity with a range of 69-100% for binary classification and range of 76-100% for multi-class classification of fibre material. Hand-held FTIR spectrometers were utilized for mid-infrared spectroscopy to examine five types of textile fibres. The Random Forest models generated using the FTIR spectra show the potential of utilising two types of reflectance interference: diffuse and specular reflectance in classifying fibre materials. Binary classification models based on specular reflectance show a prediction accuracy range of 75%-85%, sensitivity range of 88%-100% and specificity range of 86%-100%. The multiclass-classification model demonstrates a prediction accuracy range of 84%-89%, sensitivity range of 72%-100% and specificity range of 88%-100%. Binary classification models using diffuse reflectance results show a prediction accuracy range of 87%-91%, sensitivity range of 95%-100%, and specificity range of 92%-100%. The multiclass-classification model demonstrates a prediction accuracy range of 86%-90%, sensitivity range of 80%-100%, and specificity range of 93%- 100%. Gunshot residue is useful evidence that can be used to identify ammunition type and calibre. Five different ammunition types have been studied here, each with various calibres, in relation to textile fabrics that have been used previously in this study. The similar types were studied using microscopic ATR-FTIR and a classification algorithm to build random forest models. Also, the microscope view shows the difference in morphology information about the particle for the colour and shape. The classification models in the binary and multi-class classification for FTIR data used here achieved 100% for accuracy, sensitivity, and specificity.
Advisor / supervisor
  • Palmer, David (David S.)
  • Baker, Mathew J.
Resource Type
Note
  • Previously held under moratorium from 29th August 2024 until 29th August 2026.
DOI
Funder
Embargo Note
  • This thesis is restricted to Strathclyde users only until 29th August 2029.

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details File 2024-09-05 Embargo