Total monomeric anthocyanin concentration estimation of red wines based on color data using multivariate statistical analysis and machine learning

BACKGROUND Red wine is a high‐value grape derivative containing complex chemical compounds that dictate its quality. Qualitative and quantitative determination of these compounds is essential for standardizing and classifying the product. This proof‐of‐concept study integrates chemometric methods and machine learning algorithms to establish predictive models for estimating total monomeric anthocyanin (TMA) concentration in red wines using non‐destructive colorimetric data. RESULTS Images were generated and analyzed using wine samples at three liquid depths (2, 3 and 4 mm) across three color spaces (RGB, HSV and Lab). To predict TMA content from extracted digital image features, partial least squares regression (PLSR), random forest (RF) and artificial neural network (ANN) models were evaluated using external validation and stratified five‐fold cross‐validation. Optimal imaging configurations were identified as 2 mm + HSV for PLSR and 2 mm + Lab for both RF and ANN. Cross‐validated performance metrics yielded mean test R 2 values of 0.897 ± 0.019 for PLSR, 0.960 ± 0.011 for RF and 0.973 ± 0.020 for ANN. Corresponding root mean square errors were 0.252 ± 0.032, 0.174 ± 0.023 and 0.113 ± 0.037, respectively. A one‐way analysis of variance confirmed significant differences among the models ( P < 0.05), and Tukey's honestly significant difference grouped the non‐linear RF and ANN models together, separating them from PLSR. CONCLUSION Under these controlled experimental conditions, non‐linear methods (RF and ANN) demonstrated significantly stronger predictive performance than the linear PLSR model ( P < 0.05). This methodology offers a rapid, non‐destructive and highly promising approach for the accurate determination of TMA.