Spectral transformation and wavelength selection in near-infrared spectra classification

Near-infrared spectroscopy is used to discriminate between different dosage strengths of tablets in blister packs. Univariate and multivariate feature selection and three data transformation methods are studied. The second derivative appears to be the most effective transformation. Univariate feature selection methods enable the selection of two selective bands, the difference of which can directly separate all groups. The bipolar axis in the biplots of scores and loadings obtained with PCA or PLS also enable the selection of discriminating ratios of wavelengths.