Learning Wine Quality

March 2018

Coursework for EE3-23 Introduction to Machine Learning,
Imperial College,
London, United Kingdom

In this investigation I sought to develop a prediction model of quality scores for red and white wines. For each class of wine, three types of linear regression were generated: ordinary least squares regression (used as a baseline predictor), linear regression using regularization and regression using a support vector machine. For each regression technique, the model’s hypotheses were cross-validated and the best hypothesis was selected by empirical risk minimization. For both classes of wine, I observed that regression using a support vector machine with a linear kernel yielded the best error performance.