Detailed Study of Wine Dataset and its Optimization

Abstract

The consumption of wine these days is becoming more common in social gatherings and to monitor the health of individuals it's very important to maintain the quality of the wine. For the assessment of wine quality many methods have been proposed. We have described a technique to pre-process the “Vinho Verde” wine dataset. The dataset consists of red and white wine samples. The wine dataset size has been reduced from a total of 13 attributes to 9 attributes without any loss of performance. This has been validated through various classification techniques like Random Forest Classifier, Decision tree Classifiers, K-Nearest Neighbor Classifier and Artificial Neural Network Classifier. These classifiers have been compared based on two performance metrics of accuracy and RMSE values. Among the three classifiers Random Forest tends to outperform the other two classifiers in various measures for predicting the quality of the wine.