Simmons, Rory (2021) The application of multivariate analysis to aid interpretation of textile fibre dyes analysed by microspectrophotometry. Doctoral thesis, Northumbria University.
|
Text (Doctoral Thesis)
simmons.rory_phd_14030299.pdf - Submitted Version Download (5MB) | Preview |
Abstract
There have been numerous recent publications calling for an increase in the reliability of forensic evidence. Furthermore, there have been comments on a noticeable lack of research published with regard to the application of multivariate analysis to textile fibre evidence.
In this work a classification system was proposed that would utilise a probabilistic approach, require minimal user input, and be robust. The system utilised microspectrophotometry data collected from various fibres - without the use of additional analytical techniques such as microscopy or thin layer chromatography to represent a more streamlined and objective approach. A set of optimal settings for the classification system were established through experimentation utilising acrylic and cotton fibres from indistinguishable and distinguishable sources. In addition, two multivariate analysis approaches were investigated; the application of principal component analysis for dimension reduction followed by linear discriminant analysis (PCA-LDA) and the utilisation of linear discriminant analysis alone (LDA-own). The optimal settings for the proposed classification system were found to be upper/lower self-predictive probability (SPP) = 0.9999/0.0001, exceedance proportion (EP) = 0.5 and number of fibres per group = 10.
Up to 100% classification accuracy was observed when considering both fibres from indistinguishable and distinguishable sources – provided that 10 fibres were available from both sources and that the dye composition of both sources were suitably dissimilar if they were truly from different sources. If only single fibres were available for analysis, or the dye composition between truly different sources of fibres was too similar then classification accuracy decreased.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | chemometrics, principal component analysis (PCA), linear discriminant analysis (LDA), forensic science, machine learning |
| Subjects: | F100 Chemistry F200 Materials Science |
| Department: | Faculties > Health and Life Sciences > Applied Sciences University Services > Graduate School > Doctor of Philosophy |
| Depositing User: | John Coen |
| Date Deposited: | 26 Mar 2021 09:40 |
| Last Modified: | 10 Jun 2022 08:01 |
| URI: | http://nrl.northumbria.ac.uk/id/eprint/45795 |
Actions (login required)
 |
View Item |
Downloads
Downloads per month over past year
