CLASSIFICATION OF WINES BY APPLYING PATTERN RECOGNITION TO CHEMICAL COMPOSITION DATA

Abstract

ABSTRACT Pattern recognition techniques are applied to the classification of wines by their chemical compositions. A study has been completed on 49 samples of 1970 and 1971 vintage Rhine and Moselle German white wines. The previously published data include atomic absorption analysis for elemental compositions, gas chromatographic analysis for alcohols, and the determination of total acids, solids and ash. Visual examination of the data showed no apparent separation. Several classification procedures were used and Least Squares Multilinear Classifier (LEAST) proved to be the best method for this study. Separations by vintage years and wine regions are made possible by these methods. The best chemical features used for each separation are different, which reflects the wide variations between regions and between years within the same region. Ratios of chemical features are found to be more useful for classification than values of individual measurements.