Abstract

Reliable discrimination of the geographical origin of sparkling wines is of upmost importance to ensure product quality and authenticity, impacting on product price and producer reputation. This paper proposes a multivariate-based framework aimed at identifying relevant chemical elements (features) for classifying sparkling wine samples according to the country of origin. For that matter, the framework relies on the weights generated by the reliefF algorithm as an index to assess the importance of features for sample stratification. The features with the smallest indices are iteratively eliminated, leading to the subset of features responsible for the highest classification accuracy. Hierarchical classification testing four classification techniques is used to improve multiclass categorization. A total of 111 sparkling wine samples from four countries (Argentina, Brazil, France and Spain) described by 12 chemical elements are assessed. The proposed method obtained average 100% accurate classifications when retaining only 3 of the original features: K, Li, and Mn.