Multi-class wine grades predictions with hierarchical support vector machines

Abstract

Important wine attributes found in wine reviews are used to predict a wine's grade through linear kernel support vector machines (SVMs). In this work, grade prediction is defined as a multi-class problem with four classes: 100∼95, 94∼90, 89∼85 and 84 below. Since SVMs inherently do binary classification, the multi-class problem is solved using a hierarchical approach. More than 100,000 wines are collected as our dataset. Based on the two-layer SVM model which is built in this study, we accomplish high accuracy on predicting a wine's grade. Coverage, which is usually a multi-label metric, is also adapted to evaluate these results. To the best of our knowledge, it is the first time that multi-class problem is applied to Wineinformatics.