Skip to Main Content
The results of Chilean wine classification based on the information contained in wine aroma chromatograms measured with a fast GC analyzer (zNosetrade) are reported. The aroma profiles are the results of the derivative of frequency change responses of a surface acoustic wave (SAW) detector when it is exposed to a flux of wine volatile organic compounds (VOCs) during aroma measurement. Classification is done after two sequential procedures: first applying principal component analysis (PCA) or wavelet transform (WT) as feature extraction methods of the aroma data, which results in data dimension reduction. In the second stage, linear discriminant analysis (LDA), radial basis function neural networks (RBFNNs), and support vector machines (SVMs) are used as pattern recognition techniques to perform the classification. This paper compares the performance of three classification methods for three different Chilean wine varieties (Cabernet Sauvignon, Merlot, and Carmenere) produced in different years, in different valleys, and by different Chilean vineyards. It is concluded that the highest classification rates were obtained using wavelet decomposition together with SVM with a radial base function (RBF) type of kernel.