Wine quality prediction by several data mining classification models

Abstract

People are living for a better life now, and since red wine is the symbol for a luxury life, there has been an increasing demand for good-quality red wine. Therefore, it’s essential to predict a reliable model of wine quality. This research develops a new wine quality prediction method based on the red wine data from UCI website. It focuses on using several data mining (DM) methods on various of features that are highly related to wine quality, including methods like support vector machine, random forest method, K-nearest-neighbor method and neutral network method. It also scales the data and uses PCA method to reduce data dimension and apply the methods above on the processed data respectively. By comparing properties (e.g., precision, recall, F1, error and AUC area of each model), it finally successfully predicts the most advanced classification model---the Neural network model working on the scaled data set. The model can be used to predict the taste preferences and can help producers to enhance the red wine taste and quality. Since the model eliminates the influence of unimportant features, it is more reliable. These results shed light on the evaluation during wine production.