Development of a fast and reliable method for long- and short-term wine age prediction

Wine age prediction based on its intrinsic characteristics can provide significant assistance to oenologists' quality evaluations, concerning wine ageing process control and wine quality assurance. Simpler, faster, cheaper and affordable analytical procedures would be greatly welcome to establish such a practice. In this study, we present a new and reliable strategy to predict wine age, in the long and short-term, centered on the use of wine UV-vis absorbance data, coupled with proper chemometric techniques. The strategy followed consists essentially in first pre-processing the UV-vis data, secondly to carry out variable selection over such pre-processed data sets, and finally to use the set of selected variables for developing a PLS model focused on wine age prediction. We tested different data pre-processing methodologies, namely first and second derivatives, multiplicative scatter correction, standard normal variate and orthogonal signal correction, as well as different variable selection approaches, specifically interval partial least squares, VIPS, genetic algorithms and the wavelet transformation combined with a genetic algorithm. In both case studies, regarding long and short-term ageing periods, we have found out that it is indeed possible to predict wine ages, in our case Madeira wine ages, with an accuracy of 1.4 years for longer ageing periods, and of 3 months for wines of an age comprised in the first two years of ageing. The genetic algorithm revealed to be very useful for proper wavelet coefficients selection, leading to the most parsimonious model among all those analyzed, which also presents the best predictive performance found.