Typification of vinegars from Jerez and Rioja using classical chemometric techniques and neural network methods

Abstract

This study demonstrates that it is possible to characterise the vinegars obtained from wines with Certified Denomination of Origin Rioja (66 vinegars) and Jerez (18 vinegars) according to their chemical composition. SIMCA was used, along with cross-validation, as a modelling multivariate technique. In order to demonstrate that no better sensitivity and specificity of SIMCA can be achieved, a comparison was made with the results obtained by using GINN, which is a neural network with stochastic learning, which directly optimises both parameters. It was found that 92.9% of the classifications obtained by cross-validation with SIMCA were accurate, whereas with GINN 88.7% were correct (median of 10 training steps). The sensitivity and specificity obtained with SIMCA were up to 85% for Rioja vinegars and 95% for Jerez vinegars. The neural network gives higher values than those mentioned above for these parameters which confirms their optimal character. The modelling and discriminant capacities of the 20 chemical variables were also studied.