Comparative study of artificial neural network and multivariate methods to classify Spanish DO rose wines

Area of Science:

• * Enology and Wine Chemistry
• * Chemometrics and Analytical Chemistry

Background:

• * Geographical origin is a key factor in wine quality and consumer perception.
• * Objective classification of Spanish rose wines (Denominación de Origen - DO) is crucial for authenticity and market differentiation.

Purpose of the Study:

• * To apply and compare multivariate analysis and artificial neural networks for classifying Spanish rose wines based on geographical origin.
• * To identify key chemical variables that effectively differentiate wines from distinct Spanish DO regions.

Main Methods:

• * Analysis of 70 commercial rose wines from four Spanish DOs (Ribera del Duero, Rioja, Valdepeñas, La Mancha) across two vintages.
• * Measurement of 19 different wine variables.
• * Application of Stepwise Linear Discriminant Analysis (SLDA) and Artificial Neural Networks (ANN) for classification.

Main Results:

• * SLDA model selected 10 variables, achieving 98.8% correct classification and 97.3% global prediction.
• * ANN model selected seven variables (five overlapping with SLDA) and reached 100% correct classification for both training and prediction.
• * Selected variables provided enologically interpretable information for wine differentiation.

Conclusions:

• * Both SLDA and ANN are highly effective and acceptable methods for classifying Spanish rose wines by geographical origin.
• * The identified variables are robust indicators for differentiating wines based on their DO.
• * Chemometric approaches offer valuable tools for enological analysis and wine authentication.