Total monomeric anthocyanin concentration estimation of red wines based on color data using multivariate statistical...

Area of Science:

• Food Science
• Analytical Chemistry
• Data Science

Background:

• Red wine quality is determined by complex chemical compounds.
• Accurate quantification of these compounds is crucial for standardization and classification.
• Non-destructive methods are sought for wine quality assessment.

Purpose of the Study:

• To develop predictive models for estimating total monomeric anthocyanin (TMA) concentration in red wine.
• To utilize non-destructive colorimetric data and chemometric methods.
• To compare the performance of different machine learning algorithms for TMA prediction.

Main Methods:

• Digital images of wine samples were captured at varying liquid depths (2, 3, 4 mm) and analyzed across RGB, HSV, and Lab color spaces.
• Partial Least Squares Regression (PLSR), Random Forest (RF), and Artificial Neural Network (ANN) models were employed to predict TMA content.
• Model performance was evaluated using external validation and stratified five-fold cross-validation.

Main Results:

• Optimal imaging configurations were identified: 2 mm + HSV for PLSR, and 2 mm + Lab for RF and ANN.
• Non-linear models (RF and ANN) showed superior predictive performance compared to the linear PLSR model.
• RF achieved an R² of 0.960 ± 0.011, and ANN achieved 0.973 ± 0.020, with lower root mean square errors.

Conclusions:

• Non-linear machine learning models (RF, ANN) significantly outperform linear models (PLSR) for predicting TMA in red wine.
• This methodology provides a rapid, non-destructive, and accurate approach for determining TMA concentration.
• The findings support the use of colorimetric data and chemometrics for objective wine quality assessment.