Wine authentication: a fingerprinting multiclass strategy to classify red varietals through profound chemometric analysis of volatiles

Abstract

The verification of the grape variety with chemical–analytical methods is one of the major challenges in wine authentication. Such strategies use multivariate data analysis and are expected to separate individual grape varieties; also, the classification models for a large number of varieties shall give accurate predictions. In the part II of a non-targeted fingerprinting study presented herein, special multiclass chemometric strategies for the classification of German and non-German red wine varieties available on the German market were tested. The obtained three-dimensional raw data of a standardised headspace solid phase microextraction (HS-SPME) online coupled with gas chromatography mass spectrometry (GC–MS) was used; a metabolomics software and data pre-treatment were applied. The feasibility of the approaches was determined with four botanical origins by testing the models with external samples (validation). In particular, suitable modelling of similar wine varieties was a discriminant strategy using one-versus-one models based on orthogonal partial least squares discriminant analysis under the direction of a decision tree: on average, 85–98% correct classification of external test samples through ten tests was achieved. In addition, soft independent modelling of class analogies confirmed the classification. Both statistical strategies may be recommended for further improving wine authentication.