Modeling the Preference of Wine Quality Using Logistic Regression Techniques Based on Physicochemical Properties

Abstract

Wine quality is attributed to many different factors of the wine working collectively to bear a sensory experience that is not apparent from considering these components in isolation. The various chemical components in wine give the wine its distinct taste and aroma. Appreciation of wine quality involves moving beyond our innate preferences. Currently, about 1300 components relating to wine quality have been identified in wine and new components continue to be found. These physicochemical properties can be used to model wine quality. This review presents an analysis to extend what P. Cortez, A. Cerdeira, F. Almeida, T. Matos and J. Reis accomplished using support vector machine and neural network methods for modeling wine preferences by data mining from physicochemical properties. Two logistic regression approaches are used to predict human wine taste preferences with the goal of better predictions. The data were subject to the logistic regression analysis to develop suitable equations to predict which components were significant in the determination of quality of wine. Since ordering exists in the dependent variable, we first considered using ordinal logistic regression. Ordinal logistic regression is a statistical technique whose dependent variable is the order response category variable and the independent variables may be categorical, interval or ration scale. An order response variable is useful for subjective assessment of quality, importance or relevance. After applying this technique, we realized that sulphate, which improves the scent of wine, and citric acid were significant as an indication of quality in both red and white wine. As some of the assumptions of the ordinal logistic model were violated, we employed multinomial logistic regression as well. Multinomial logistic regression is used when the dependent (response) variable in question is nominal, i.e. a set of categories which cannot be ordered in any meaningful way (for example, societal class) and consists of more than two categories. It assum