Classification and prediction of tannin botanical origin through voltammetry and machine learning approach

Abstract

The classification of enological tannins has gained importance following the OIV’s requirement to include their botanical origin on product labels (OIV-OENO 624-2022). A rapid classification method would be particularly valuable for producers and retailers, enabling them to quickly determine the origin of tannins. This study explores a novel approach using linear sweep voltammetry (LSV) coupled with machine learning algorithms to classify enological tannins. While traditional methods such as LC-MS, UV-Vis, and FTIR provide detailed chemical information, they are often time-consuming, costly, and require skilled personnel. In contrast, voltammetry offers a rapid, cost-effective alternative, albeit with challenges in interpreting the resulting voltammograms due to signal overlaps from various electrochemical processes. Interpreting these results may require advanced data processing, such as signal deconvolution (Ugliano, 2016) and machine learning algorithms to extract insights from voltammetric patterns (Choi et al., 2022). However, the efficiency of machine learning algorithms is closely linked to a large availability of data. To address these limitations, a Generative Adversarial Network (GAN) was employed to generate synthetic voltammograms, combined with experimental data to expand the training dataset. This augmented dataset was used to train machine learning models, including Random Forest, Extreme Gradient Boosting, and Support Vector Machine (SVM), with the latter achieving the best classification results. The SVM model demonstrated high accuracy (94%) and excellent discrimination between tannin classes, as indicated by an AUC-ROC of 0.9971.

The study also integrated feature importance and Recursive Feature Elimination (RFE) analyses to identify key voltammetric features contributing to the classification. Features around 0.3 V, 0.57–0.65 V, and 1.11–1.17 V were found to be critical for distinguishing between tannin types. While the proposed method highlights the potential of combining voltammetry and machine learning for rapid tannin classification, further studies on model solutions are required to generalize the approach to different wine matrices.

This workflow provides a promising tool for the wine industry, offering a rapid, cost-effective method to classify tannins and optimize their enological applications