Accurate modelling of the performance of a yacht in varying environmental conditions can significantly improve a yachts performance. However, a racing yacht is a highly complex multiphysics system meaning that real-time performance prediction tools are always semi-empirical, leaving significant room for improvement. In this paper we first use unsupervised machine learning to analyse full-scale yacht performance data. The widely documented ORC VPP (ORC, 2015) and the commercial Windesign VPP are compared to the data across a range of wind conditions. The data is then used to train machine learning models. A number of machine learning regression algorithms are explored including Neural Networks, Random Forests and Support Vector Machines and improvements of 82% are obtained compared to the commercial tools. The use of physics-based learning models (Weymouth and Yue, 2013) is explored in order to reduce the amount of data required to achieve accurate predictions. It is found that machine learning models can outperform empirical models even when predicting performance in environmental conditions that have not been supplied to the model as part of the training dataset.