What Actually Wins Soccer Matches: Prediction of the 2011-2012 Premier League for Fun and Profit

Abstract

Sports analytics is a fascinating problem area in which to apply statistical learning techniques. This thesis brings new data to bear on the problem of predicting the outcome of a soccer match. We use frequency counts of in-game events, sourced from the Manchester City Analytics program, to predict the 380 matches of the 2011-2012 Premier League season. We generate prediction models with multinomial regression and rigorously test them with betting simulations. An extensive review of prior efforts is presented, as well as a novel theoretically optimal betting strategy. We measure performance different feature sets and betting strategies. Accuracy and simulated profit far exceeding those of all earlier efforts are achieved.