Red wine Quality Prediction by Supervised Learning

Abstract

In this essay, various machine learning techniques are carried out to model the red wine dataset, which contains 1599 samples and 11 physical features, and predict its quality. Data processing techniques including dropping duplicate values and removing outliers are first applied to normalize the features and bring them to a similar scale. Five algorithms, KNN, Multinominal Logistic Regression, Bagging, Random Forest, and Gradient Boosting are then applied. Cross-validation is performed to identify the best parameters for each model. Additionally, feature selection is also done in Logistic Regression based on the calculated p-values. All five models perform well, with test accuracies roughly around 60%, indicating that machine learning has enormous potential on red wine quality prediction to facilitate the whole industry. Among them, Logistic Regression and Random Forest have slightly higher accuracies. This paper highlights the importance of machine learning methods of Logistic Regression and Random Forest in red wine quality prediction.