Using neural networks to identify the regional and varietal origin of Cabernet and Merlot dry red wines produced in Krasnodar region

Abstract

This paper shows a possibility of establishing the authenticity and geographic origin of wines by neural networks based on multi-element analysis. The study used 144 samples of Cabernet and Merlot dry red wines produced in Krasnodar Region according to traditional technologies. The wines were provided by the producers or purchased in retail stores. The concentrations of 20 micro- and macroelements in red wines were determined by atomic emission spectroscopy with inductively coupled plasma. The analysis of average elemental contents showed a significant dependence of wine composition on the grape variety and place of origin, which enabled us to examine interrelations between the elements and think of a way to identify them by means of classification models. The software STATISTICA Neural Networks was used to assess a possibility of determining the grape variety and geographical origin. The neural networks constructed in the study contained five variables corresponding to the elements with statistically significant correlations between the names of the regions and the wine samples, namely Fe, Mg, Rb, Ti, and Na. These predictors were able to determine the grape variety and place of growth with a sufficiently high accuracy. In the test sample set, the accuracy reached 95.24% and 100% for variety and region identification, respectively. A software product was developed to automate the calculations based on the neural networks. The program can establish the grape variety from a minimal set of microelements, and then, based on the variety and the same set of microelements, determine its place of origin.