Application of chemometric analysis to UV-visible and diffuse near-infrared reflectance spectra.

Abstract

Multivariate analysis of spectroscopic data has become more common place in analytical investigations due to several factors, including diode-array spectrometers, computer-assisted data acquisition systems, and chemometric modeling software. Chemometric regression modeling as well as classification studies were conducted on spectral data obtained with chili peppers and fabrics samples. Multivariate regression models known as partial least squares (PLS-1) were developed from the spectral data of alcoholic extracts of Habanero peppers. The developed regression models were used to predict the total capsaicinoids concentration of a set of unknown samples. The ability of the regression models to correctly predict the total capsaicinoids concentration of unknown samples was evaluated in terms of the root mean square error or prediction (RMSEP). The prediction ability of the models produced was found to be robust and stable over time and in the face of instrumental modifications. A near-infrared spectral database was developed from over 800 textile samples. Principal components analysis (PCA) was performed on the diffuse near-infrared reflectance spectra from these commercially available textiles. The PCA models were combined together into a soft independent modeling of class analogy (SIMCA) in order to classify the samples according to fiber type. The samples in the study had no pretreatments. The discriminating power of these models was tested by creating validation sets within a given fiber type as well as attempting to classify samples into a category that they do not belong to. The apparent sub-class groupings within the same fiber class were investigated as to whether or not they were caused by chemical processing residues, multipurpose finishes, or dyes.