Wine quality assessment for Shiraz vertical vintages based on digital technologies and machine learning modeling.
Natalie Harris, Claudia Gonzalez Viejo, Chris Barnes, Alexis Pang, Sigfredo Fuentes
Abstract
Assessment of wine quality traits can be costly, time-consuming, and usually undertaken with complex chemical analysis and sensory evaluation in specialized laboratories. This study aimed to use novel digital technologies based on near-infrared (NIR) spectroscopy and a low-cost electronic nose (e-nose) integrated with machine learning modeling to assess wine quality traits and provenance in a vertical vintage of Shiraz wines. Results showed highly accurate machine learning (ML) models for the classification of wine vintages for Model 1 (NIR; 98.3%) and Model 2 (e-nose; 99.5%), and prediction of (i) intensity of sensory descriptors (NIR Model 3; R = 0.97), (ii) peak area of volatile aromatic compounds (NIR Model 4; R = 0.96), (iii) physicochemical parameters (NIR Model 5; R = 0.94), (iv) intensity of sensory descriptors (e-nose Model 6; R = 0.97), (v) peak area of volatile aromatic compounds (e-nose Model 7; R = 0.96), and (vi) physicochemical parameters (e-nose Model 8; R = 0.93). Winemakers may use these models to assess vintages and maintain high-quality wines associated with specific vineyards and regions or for a distinctive wine variety. Models could be developed further by including data from different vineyards, regions, wine variety, seasonality, other production, and winemaking techniques.