Classification of Spanish Red Wines Using Artificial Neural Networks with Enological Parameters and Mineral Content

Abstract

Spanish red wines from the Canary Islands were categorized into seven classes: Tacoronte-Acentejo (class T), Valle de la Orotava (class O), Ycoden-Daute-Isora (class YDI), Abona (class A), Valle de Guimar (class VG), La Gomera (class G), and La Palma (class P), and 20 samples were studied from each denomination of origin. Metal concentrations (B, Ca, Cd, Co, Cr, Cu, Fe, K, Li, Mg, Mn, Mo, Ni, Na, Pb, and Zn) and physicochemical parameters (pH, volatile acidity, total acidity, malic acid, acetic acid, reducing sugars, alcohol content, free sulfur dioxide, total sulfur dioxide, and total polyphenols) were used as descriptors to differentiate among classes. Supervised learning pattern recognition procedures were applied. Linear discriminant analysis allowed up to ~80% of correct classification. To improve discriminatory accuracy, another kind of algorithm that can model nonlinear separation among classes was considered: artificial neural networks. This method obtained excellent results, with 100% of the 140 wines correctly placed into the associated seven classes. Our results are in good agreement with the working hypothesis of differentiation among wines coming from different locations, including both different islands and different sites on Tenerife Island.