Predicting Harbertson-Adams Assay Phenolic Parameters In Red Wines Using Visible Spectra

Abstract

The Harbertson-Adams phenolic parameter assay is a well- known method to measure a panel of phenolic compounds in red wines. However, the multistep analyses required by the method fail at producing results on multiple parameters rapidly. In the present article, we analyze the bene ts of applying a statistical model based on Principal Component Analysis (PCA) and a statistical learning technique denoted as Support Vector Regression Machines (SVR) for correlating sample spectra data to the Harbertson-Adams assay, on each of the phenolics components. The resulting model showed a high correlation between the measured and predicted values for each of the phenolic parameters despite the multicollinearity and high dimensions of the dataset.