RESUMO
Discriminating wines according to their denomination of origin using cost-effective techniques is something that attracts the attention of different industrial sectors. In search of simplicity, direct UV-visible spectrophotometric techniques and different multivariate statistical techniques are used with admissible results to characterize wine produced in specific regions. However, most of the reported classification methods do not exploit all of the statistical relations in the investigated dataset and are inherently affected by the presence of outliers. The aim of this paper is to test novel classification methods such as support vector machines as a means of improving the classification rate when UV-visible spectrophotometric methods are used to discriminate wines. The advantages of such a discrimination tool are demonstrated when classification rates are compared for a large number of Spanish red and white wines and classification rates above 96% are achieved. The proposed methodology also enables the selection of the most relevant wavelengths for sample discrimination. The proposed methodology also enables the selection of the most relevant wavelengths for sample discrimination.