An Investigation into the Use of Multivariate Analysis for the Interpretation of Textile Fibre Evidence in Criminal Cases

Textile fibres are frequently encountered within forensic casework and are particularly prevalent in serious crimes such as homicide and sexual assault. Currently the interpretation of textile fibre evidence is a primarily subjective process, relying heavily on expert opinion. Although uncommon, this can lead to disagreement between experts as to whether or not a fibre(s) may have originated from a particular source. To date, limited work has been published regarding the use of multivariate analysis (MVA) to aid interpretation of textile fibres evidence; therefore this work aims to contribute to the increasing amount of information becoming available.

In this study, spectral data was obtained for 23 visually distinct cotton and acrylic sources using visible range microspectrophotometry. Linear discriminant analysis (LDA) was then performed with and without prior dimension reduction by principle components analysis (PCA). MVA was performed using the statistical software R. A blue cotton colour block study using 10 dark and 10 medium blue cotton sources was also performed. Additionally, the use of normalisation as a pre-processing method was investigated.

LDA used with and without PCA as dimension reduction demonstrated comparable results when classifying samples into their correct source; providing a suitable number of principle components were retained for subsequent analysis. Similar results were also observed when performing the colour block study using blue cotton. This suggests that the pre-processing stage of dimension reduction using PCA is not required prior to LDA when using R. Normalised and non-normalised data provided comparable results when a suitable number of principle components were included in subsequent analysis, again suggesting that this pre-process stage may not be required when performing MVA using R.