Application of Ultraviolet-Visible Absorption Spectroscopy with Machine Learning Techniques for the Classification of Cretan Wines

Abstract

The present study was aimed at the identification, differentiation and characterization of red and white Cretan wines, which are described with Protected Geographical Indication (PGI), using ultraviolet–visible absorption spectroscopy. Specifically, the grape variety, the wine aging process and the role of barrel/container type were investigated. The combination of spectroscopic results with machine learning-based modelling demonstrated the use of absorption spectroscopy as a facile and low-cost technique in wine analysis. In this study, a clear discrimination among grape varieties was revealed. Moreover, a grouping of samples according to aging period and container type of maturation was accomplished, for the first time.

A Review of Wine Authentication Using Spectroscopic Approaches in Combination with Chemometrics

“…Spectra recorded at UV and visible wavelengths (typically 190–800 nm, Figure 2 ) provide information about compounds in wine containing a chromophore, such as hydroxybenzoic (280 nm) and hydroxycinnamic (320 nm) acids, flavan-3-ols (280 nm), flavonols (370 nm), and anthocyanin glucosides (520 nm) [ 17 ]. As summarised in Table 1 , UV–Vis spectroscopy has been applied in wine discrimination according to the region of origin [ 18 , 19 ], grape variety and ageing process [ 20 , 21 ]. Although the specific chemical markers are not necessarily identified, as a non-targeted method combined with appropriate chemometric techniques such as linear discriminant analysis (LDA) and partial least squares discriminant analysis (PLS-DA), Azcarate et al were able to correctly classify Argentinian Sauvignon blanc wine samples with 100% accuracy according to their geographical origin [ 19 ].…”

“…Although the specific chemical markers are not necessarily identified, as a non-targeted method combined with appropriate chemometric techniques such as linear discriminant analysis (LDA) and partial least squares discriminant analysis (PLS-DA), Azcarate et al were able to correctly classify Argentinian Sauvignon blanc wine samples with 100% accuracy according to their geographical origin [ 19 ]. In their study, Philippidis et al achieved 97.5% correct classification of grape variety and showed that the latent variables resulting from orthogonal projections to latent structures-discriminant analysis (OPLS-DA) could be related to the absorption of aromatic compounds such as phenolic acids and flavonols [ 21 ]. In comparison to other spectroscopic methods, however, UV–Vis spectroscopy provides a limited number of spectral features; therefore, it could be used as a screening approach with more sophisticated techniques being implemented for further analysis.…”

Comparative Evaluation of Different Targeted and Untargeted Analytical Approaches to Assess Greek Extra Virgin Olive Oil Quality and Authentication

“…Both targeted and untargeted techniques have been previously combined with machine learning methods, such as PLS-DA, SVMs, Neural Networks etc. Indicatively, various agricultural products such as honey [19,20] and wine [21][22][23][24][25][26] have been investigated, considering a wide range of issues and authentication purposes such as their quality, variety and geographical and botanical origin.…”

Combination of Machine Learning and Analytical Correlations for Establishing Quantitative Compliance between the Trolox Equivalent Antioxidant Capacity Values Obtained via Electron Paramagnetic Resonance and Ultraviolet–Visible Spectroscopies

“…Chemical analysis in wine studies is an exemplified case in this field. Machine learning supports analysis of wine authenticity, quality, geographical origin, and classification on the base of spectroscopic and chemical analytic properties of samples [ 14 , 15 , 16 , 17 , 18 ]. At the same time, a majority of these methods involve processing extensive datasets to simply predict a desired output in a “black box” manner.…”