Predicting Outcomes of Association Football Matches Based on Individual Players' Performance
Abstract
This master's thesis concludes our five years study in Computer Science, at the Norwegian University of Science and Technology.Predicting the outcome of football matches is a research area where it is possible to earn a lot of money, if the generated predictions are accurate enough. In this thesis we develop three prediction models, based on a model proposed by Rue and Salvesen. Our models are scaled versions of the original model, where the scaling factors are determined by the strength of the players participating in a match. They are modelled as Bayesian networks, where the predictions are found by the Markov chain Monte Carlo method Gibbs sampling. The models are applied to the betting market for three seasons, using three different betting strategies, along with the unscaled Rue and Salvesen model. Over these three seasons, our best model, the GoalScaled model, is able to outperform the baseline Rue and Salvesen model and earn money in all seasons.