Independent components analysis as an alternative to principal component analysis and discriminant analysis algorithms in the processing of spectrometric data

Abstract

The possibility of the application of independent component analysis (ICA) to searching patterns in the spectrometric datasets and to discriminating objects is demonstrated. The data of XRF analysis of base enamels, IR spectra of automotive lacquers, and 1H NMR spectra of wines from different regions of Germany are selected for the study. In all three cases, ICA reliably separates groups of objects, increasing the percentage of correct predictions for new samples not included into the model. Moreover, ICA gives results comparable with specialized discriminant analysis methods (linear discriminant analysis, projections to latent structures discriminant analysis, and factorial discriminant analysis) in the classification of the NMR spectra of wines.