Application of multivariate regression methods to predict sensory quality of red wines

Several multivariate methods including partial least squares (PLS) regression, principal component regression (PCR) or multiple linear regression (MLR) have been applied to predict wine quality, based on the definition of chemical and phenolic parameters of grapes and wines harvested at different ripening levels. Three different models including grape phenolic maturity parameters (grape), wine phenolic parameters (wine) and a combination of grape and wine phenolic parameters (grape + wine) were analysed for each of the wine sensory attributes. The grape parameter model has been presented as the best test to predict wine quality based on sensory scores. On the other hand, wine models showed lower accuracy. The combination of grape and wine parameters presented intermediate results showing sometimes good predictability. Moreover, PLS and PCR appeared as more accurate multivariate methods compared to MLR. Although MLR showed higher correlation coefficients, lower RPD values were observed, displaying thus its lower prediction accuracy. Multivariate calibration statistics appeared as a promising tool to predict wine sensory quality in an easy and inexpensive way.