Beating the bookmakers - Using artificial neural networks to profit from football betting
Abstract
Artificial Neural Networks (ANNs) have throughout the years been used for several different purposes. Problems spanning from image classification to text generation have all been subject to ANNs.
In this report, ANNs were used in order to predict the outcomes of football matches. Using data from the football statistics web site www.whoscored.com, ANNs were constructed in order to predict the outcomes of matches from two successive seasons of the English Premier League. The predictions were then used to decide whether or not to place bets on the outcomes, in an effort to generate a profit.
Several ANNs were constructed, utilizing data sources from player ratings to team characteristics. The networks were trained using simple back-propagation training. The predictions were then used together with odds from seven international bookmakers, trying to generate a profit from betting. Different money management (betting) strategies were applied, in order to highlight the importance of choosing correct bet sizes.
The results show that simple assumptions may be enough to accurately predict the outcome of a football match. The results also show that ANNs can indeed beat bookmakers in their own game, and gain a profit from football betting.
The report ends with the author's thoughts on how to further improve the profitability of the presented models.