This paper presents a study on predicting football match outcomes using various machine learning models, aiming to outperform traditional bookmakers. The primary objectives were to develop predictive models, identify key factors influencing match results, and assess the potential monetary value of the models. The study utilized an ANOVA test for feature selection, revealing that differences in team quality metrics, such as Elo ratings and FIFA player ratings, are among the most significant predictors. Despite achieving competitive accuracy, with Linear Discriminant Analysis (LDA) reaching 68.8 %, the models generally underperformed compared to bookmakers' odds, which also achieved 68.8 % accuracy. Betting strategies were tested over 109 matches, where the probability betting strategy yielded profits for all models, with LDA achieving a 21.5 % profit, slightly surpassing the bookmakers' strategy. However, the expected value betting strategy resulted in losses, indicating a challenge in predicting non-favorite outcomes. The findings suggest that while the machine learning models developed in this project show promise in predicting match results, they require further refinement to consistently outperform traditional bookmakers, especially in identifying valuable betting opportunities.
| Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:uu-532610 |
| Date | January 2024 |
| Creators | Karlsson, Fabian, Vigholm, Albin |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
| Format | application/pdf |
| Rights | info:eu-repo/semantics/openAccess |
| Relation | MATVET-F ; 24024 |
Page generated in 0.0015 seconds