Advancing wine authentication: non-invasive near-infrared spectroscopy and machine learning for vintage and quality traits assessment

Abstract

Wine fraud, encompassing counterfeiting and adulteration, poses a significant threat to the wine industry, resulting in annual losses totalling billions of dollars. Therefore, the development of rapid, accurate, and non-invasive techniques for detecting fraud-related indicators such as wine provenance and quality traits is critical. This study aimed to develop a novel non-invasive approach utilizing a handheld near-infrared (NIR) spectroscopy device to analyse wines through their bottles. Machine learning models were developed for Shiraz wines from an Australian winery using the NIR absorbance values to predict the wine vintage (Model 1) as a classification model and the intensity of sensory descriptors (Model 2) as a regression model. The models had high accuracies of 97% for Model 1 and an R = 0.95 for Model 2. This method offers a cost-effective and rapid solution for winemakers and retailers to evaluate wine quality and authenticity without the need to open bottles. Furthermore, it holds potential for expanding assessments to include additional authentication criteria such as region, country of origin, and grape variety, thus enhancing the integrity of the wine market.