| dc.description.abstract | Traditionally, to make money on sports betting, one should have a high knowledge about the teams involved, and some luck. With the introduction of online betting exchanges, one can take the role of a bookmaker and offer odds to other players. If betting at high odds while offering someone else lower odds, one is guaranteed to win, nomatter if the team wins or loses. Before a game starts, the odds are fluctuating, and one can make money if one can predict the direction of the changes.This thesis discusses methods to predict odds changes in order to make sure wins before the game starts. Five theories about odds fluctuation are discussed and tested, and C4.5 decision trees and linear regression models are made based on the results of these tests and previously recorded odds fluctuations. A betting robot is made to bet with real money, using these models.British and Norwegian soccer are used as examples in this thesis, but the system can also be used with other sports as long as there are enough previously recorded odds to use as training data. When tested with British soccer, where 20-30 games were available for training data for each team, the system gave a 4% profit. The results depend heavily on enough training data, shown by a 3%loss when tested on Norwegian soccer, where only 10-12 games are used to train the models. Although test result show that some profit was made on test games, they also show that the variance is quite high and that one must expect to lose some times, meaning the luck aspect is not totally gone. | nb_NO |
