Application of FT-IR and machine learning in the characterization of organic and biodynamic wines vs conventional wines

Abstract

In this study, Fourier-transform infrared spectroscopy (FT-IR) combined with machine learning (ML) algorithms was employed to characterize wines according to the specific viticulture and winemaking practices (conventional, organic, and biodynamic). A total of 83 wine samples, including both red and white varieties, were analyzed. FT-IR spectral data were processed using unsupervised methods, such as principal component analysis (PCA), and supervised classification algorithms, including random forest (RF), support vector machines (SVM), and linear discriminant analysis (LDA). PCA showed a clear distribution between white and red wines, although no evident clustering was observed based on specific vineyard and winemaking practices. Conversely, the application of supervised algorithms (SVM and RF) to the FT-IR data enabled accurate classification of white wines according to their production method, achieving 100 % accuracy on the test set for both models. In the case of red wines, the SVM model yielded the best performance, with an accuracy of 83.33 % in the test set. These results highlight the potential of FT-IR spectroscopy combined with ML as a reliable approach for determining the production method of wine samples. Furthermore, a web-based platform has been developed to facilitate access to the implemented classification algorithms.