Machine learning-assisted spectroscopic methods for detecting adulteration in Barrantes wine from Folla Redonda grapes

Abstract

The present study explores the application of advanced machine learning algorithms combined with vis-NIRS and FTIR spectroscopy to detect and quantify adulteration in Barrantes wine, produced from the Folla Redonda grape, a variety exclusive to the Galicia region in Spain. Given the unique characteristics of Folla Redonda grapes, such as their distinctive phenolic profile and high antioxidant content, Barrantes wine is at risk of adulteration with lower-quality red wines. The study employs a variety of preprocessing techniques, including normalization and derivative transformations, and evaluates classification models such as Support Vector Machines (SVM), Partial Least Squares Discriminant Analysis (PLS-DA), and Random Forest (RF) for detecting adulteration at varying levels (10 % – 60 %). The findings show that SVM with SNV preprocessing provides the highest accuracy for vis-NIRS and FTIR data (100 %). Furthermore, regression models, particularly SVMRadial, were used to predict the level of adulteration with high precision (R2 > 0.95; RMSE <11 %). This research is pioneering in its field, as it is the first to optimize machine learning techniques for adulteration detection in wine. The findings provide a valuable tool for quality control in the agro-food sector, specifically for niche wines like Barrantes. The results of this study underscore the potential for rapid, precise, and non-destructive analysis through the integration of these approaches.