Neural networks and ft-ir spectroscopy for the discrimination of single varietal and blended wines. A preliminary study.

Abstract

Blending wines from different grape varieties is often used in order to increase wine complexity and balance. Due to their popularity, several types of blends such as the Bordeaux blend, are protected by PDO legislation. In the case of monovarietal wines blending is forbidden, however there is no method to authenticate their status, and for this reason adulteration can are difficult to identify. Fourier Transform Infrared Spectroscopy (FT-IR) has proven successful for the discrimination of wines based on several parameters such as geographical origin and type of aging[1], while the use of Neural Networks is now used more often for the development of prediction models. FT-IR spectroscopy coupled with Neural Networks have been used to develop a prediction model for the discrimination of single varietal and blended wines. Generalized RSquare for the training set was 0,9011 and 0,689 for the validation set, while the -Loglikelihood was 3,918 for the training and 0,111 for the validation set. The misclassified rate was 0,03 for the training set and 0,11 for the validation set, showing very good potential for the use of IR spectroscopy for the authentication of single varietal and blended wines.