Automatic classification of infrared spectra using a set of improved expert-based features Original Research Article

Three types of spectral features derived from infrared peak tables were compared for their ability to be used in automatic classification of infrared spectra. Aim of classification was to provide information about presence or absence of 20 chemical substructures in organic compounds. A new method has been applied to improve spectral wavelength intervals as available from expert-knowledge. The resulting set of features proved to be better than features derived from the original intervals and better than features directly derived from peak tables. The methods used for classification were linear discriminant analysis and a back-propagation neural network; the latter gave a better performance of the developed classifiers.