Classification of Document Papers by Infrared Spectroscopy and Multivariate Statistical Techniques

Abstract

Infrared (IR) spectra of different varieties of document papers were collected with the use of attenuated total reflectance (ATR, 4000-650 cm−1, eight paper varieties) and diffuse reflectance (DRIFTS, 9000-2500 cm−1, six paper varieties) techniques. The spectral data were classified by the application of soft independent modeling of class analogies (SIMCA), using principal components analysis (PCA) to estimate the distance of separation between the different classes of paper samples and discriminant analysis (DA) to obtain a probabilistic classification. The use of DA on spectral data needed a preliminary data reduction step, either by PCA-decomposition of spectra or the selection of discrete spectral features having maximum discriminating ability. The aim of this research was to evaluate these data-reduction techniques and compare the discriminating power of these two spectral techniques (DRIFTS and ATR) by the application of PCA and DA. The use of PCA scores as DA variables provided the best resolution (100% correct classification) for the DRIFTS spectra, while PCA on the ATR spectra resulted in the best discrimination, separating 67.86% paper pairs completely with the use of cross-validation. The results of this study reemphasize that infrared spectroscopy coupled with multivariate statistical methods of analysis could provide a powerful discriminating tool for the forensic questioned document examiner.