The Relationship Between Physicochemical Properties And...

Introduction and Motivation There is a big wine market in the world, as it plays a pivotal role in many social gatherings. Because of this, it is absolutely essential for the wine industry to be able to determine what physicochemical properties are essential for wine to be given a good rating and overall, increase its market value. Hence there is a need to investigate the influence of these properties for both wine manufacturing and selling purposes. The relationship between physicochemical properties and sensory analysis is not easy to understand[3], which makes wine classification a difficult task, as rating is based mainly on taste preference. Hence we were motivated to come up with a model that could predict wine preferences solely†¦show more content†¦The physicochemical tests include 11 continuous variables, such as determination of alcohol, sulfates or pH values. Research Questions The goal of this paper is to classify wine quality based on physicochemical and sensory analysis. For this purpose, this study uses the Neural Nets toolbox, SVM (Support Vector Machine) library, and Logistic Regression in Matlab, and Multivariable Regression in R to analyse and classify wine quality. This paper determines the most influential features for red wine quality ratings and answer the question of which machine learning algorithm performs the best in classifying wine quality. Experiment 4.1 Neural Nets We are using the Neural Nets Toolbox which is already implemented in Matlab to check the accuracy of quality classification for our wine data set. To use this toolbox, we needed to input the training data matrix, which is a 1599x11 matrix (with 11 features) and a target matrix made up of our quality outputs ranging from 3 to 8. The target matrix has the same number of rows as there are samples of wine. Hence it has 1599 rows. Each row gives the binary label vector for one of the classifications. The number of columns depends on how many classes there are. Each column representing one class. For each wine sample, if the class is n, the nth column would have a value of 1 and every other column would have a value of 0. So if there were 6 classes (as quality ratings ranges from 3 to 8), for