This application note details the use of bagging (random forest, RF) and boosting (extreme gradient boosting, XGBoost) machine learning (ML) algorithms to predict red wine mouthfeel attributes from simple chemical measurements. A panel of 15 wine experts assessed 30 commercial red wines (Australian and Spanish) using rate-all-that-apply sensory methodology and chemical data, including linear sweep voltammetry, excitation emission matrix, and absorbance, were collected. Principal component analysis simplified the sensory data, revealing four independent mouthfeel dimensions: 'drying', 'full body', 'velvety', and 'gummy'. RF and XGBoost models outperformed classical partial least squares regression, achieving over 80% accuracy on test data. This study demonstrates the potential of ML coupled with simple chemical measurements for rapid and cost-effective prediction of wine sensory properties.
A-TEEM Spectroscopy
HORIBA제품의 자세한 정보를 원하시면, 아래의 양식에 내용을 입력을 부탁드립니다.
