Wine is one of the world's higher value agricultural products. The present work is centred on Madeira wine, a fine and prestigious example among Portuguese liqueur wines,with the main goal to deepen our understanding of relevant phenomena going on during the winemaking process, in particular during ageing of "Malmsey" Madeira wine. In this paper we present the results obtained from the chemical characterization of how its aroma composition evolves during ageing, and the development of a robust framework for analyzing the identity of aged Madeira wines. An extended ageing period was considered, covering a time frame of twenty years, from which several samples were analyzed in terms of their aromatic composition. The multivariate structure of this chemical information was then processed through multivariate statistical feature extraction techniques such as principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA), in order to identify the relevant patterns corresponding to trends associated with wine ageing. Classification methodologies for age prediction were developed, using data from the lower dimensional sub-spaces obtained after projecting the original data to the latent variable spaces provided by PCA or PLS-DA. Finally, the performance for each classification methodology developed was evaluated according to their error rates using cross-validation methodologies (Leave-One-Out and k-fold Monte Carlo). Results obtained so far show that quite interesting classification performances can indeed be achieved, despite the natural variability present in wine products. These results also provide solid bases which can be used to build up available frameworks which assist quality monitoring and identity assurance tasks.
Copyright 2009 Elsevier B.V. All rights reserved.