The goal of this project was to develop a supervised machine learning algorithm, which could enable video games to adapt to the preferences of the individual player. This had been done before using unsupervised techniques to develop the player behaviour models. However, these models were being difficult to use in the design process since the mappings were hidden. In this research we used the Core Game Dynamics (CGD) model and questionnaire to model player behaviour. We developed a game based on the CGD model, that would appeal to both the Assault, Manage, and Journey player types. We then gathered data from 102 participants, and used this data to train a supervised learning classifier offline. We then evaluated this classifier and found it had a general accuracy of 48%. Furthermore, we tried splitting the data into different segments with different characteristics. Here we found that segmenting players by the mean difference between player types had a positive impact on performance, achieving an accuracy of 71%.