Piergiorgio Comuzzo_Universidad de Udine (Italia) 

Cyclic Voltammetry (CV) is a powerful, rapid and versatile tool in wine analysis, enabling the characterization of polyphenols and antioxidants, monitoring of oxidation and ageing, estimation of reactive components such as SO₂, and providing fingerprints for classification and ageing studies. In the present research, CV was applied to study the evolution of thirty-six white wines during bottle ageing. Wines were supplied by local producers after bottling, stored in controlled conditions (20±5 °C and 70±10 % RH) and analyzed after two, six, ten and twelve months. In addition to CV, analyses consisted of seventy-seven analytical parameters relevant for wine evolution and ageing, including free and total sulfur dioxide, flavanols, carbonyl and volatile compounds, a browning assay (POM-test), wine color and total phenolics. Data were processed by stepwise regression, an automated feature selection method in machine learning and statistics that iteratively adds or removes predictor variables to build the best-fit regression model. Using this algorithm, we identified a parsimonious subset of voltammetric data showing conditional associations with other analytical variables, evaluating which potential/current coordinates in the voltammogram best predict how these variables changed over time during storage. Regression models were optimized for each analytical parameter; few points (max thirteen) of the voltammograms collected after two months of ageing, were generally able to predict the values of different analytical parameters after twelve months, with the best fitting obtained for flavanols. Results demonstrate that CV can be used for predicting the ageing potential of a wine, allowing winemakers to decide more knowingly the strategies for stabilizing their products at bottling, modelling their shelf-life. This study shall be improved increasing the sample size and the variability of the wines, as well as testing more powerful machine learning predictive tools, such as neural networks.

Piergiorgio Comuzzo(1), Angelo Topo(1), Rosanna Toniolo(1), Roberto Pagliarini(2)
(1)Università degli Studi di Udine, Dipartimento di Scienze Agroalimentari, Ambientali ed Animali, via Sondrio 2/A, 33100 Udine – Italy
(2)Università degli Studi di Udine, Dipartimento di Scienze Matematiche, Informatiche e Fisiche, via delle Scienze 206, 33100 Udine – Italy