Beating the Bookies
(2024) STAN40 20241Department of Statistics
- Abstract
- In this study, we have explored the viability of using machine learning models to predict the outcomes of Premier League football matches. Especially we have evaluated the effectiveness with the Kelly Criterion as a betting strategy. We processed a comprehensive dataset to develop six distinct predictive models, each designed to forecast the probability of three possible match outcomes: a home win, a draw, or an away win. These models included various advanced machine learning techniques, and their performance was enhanced through hyperparameter tuning employing both grid search and randomized search methods.
Our methodology involved simulating an entire Premier League season, applying the Kelly Criterion to manage and allocate bets... (More) - In this study, we have explored the viability of using machine learning models to predict the outcomes of Premier League football matches. Especially we have evaluated the effectiveness with the Kelly Criterion as a betting strategy. We processed a comprehensive dataset to develop six distinct predictive models, each designed to forecast the probability of three possible match outcomes: a home win, a draw, or an away win. These models included various advanced machine learning techniques, and their performance was enhanced through hyperparameter tuning employing both grid search and randomized search methods.
Our methodology involved simulating an entire Premier League season, applying the Kelly Criterion to manage and allocate bets based on the probabilistic outputs of our models. The objective was to evaluate the potential returns on investment from each model. Among the models tested, the neural network emerged as the most successful, yielding a return of 35.48 times the initial bankroll. Other models demonstrated varied levels of success, illustrating the diverse potential of machine learning applications in sports betting.
The results of our simulations suggest that while machine learning can indeed be a powerful tool in sports betting, its efficiency depends on several factors. The quality of the input data and the level of sophistication of feature engineering might influence model performance. Our study highlights that further enhancements, such as incorporating more information and/or variables in our data and exploring alternative approaches to feature engineering, could improve predictive accuracy. Additionally, domain knowledge remains a critical component, suggesting that a hybrid approach combining data-driven techniques with expert insight may yield the most robust long-term betting strategies.
In conclusion, while our findings are promising, indicating that the application of machine learning and the Kelly Criterion can be profitable, they also underscore the need for ongoing refinement and the integration of comprehensive domain knowledge to fully capitalize on this approach in the competitive sports betting market. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9158848
- author
- Båth Viderström, Oscar LU and Sjöberg, Viktor LU
- supervisor
- organization
- alternative title
- Harnessing Machine Learning and the Kelly Criterion for Premier League Betting Profits
- course
- STAN40 20241
- year
- 2024
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- Premier League, Betting, Machine learning, Neural Network, XGBoost, Random Forrest Classifier, Naive-bayes, Support Vector Machine, Logistic Regression, Kelly Criterion, Data Science.
- language
- English
- id
- 9158848
- date added to LUP
- 2024-06-17 14:23:55
- date last changed
- 2024-06-17 14:23:55
@misc{9158848, abstract = {{In this study, we have explored the viability of using machine learning models to predict the outcomes of Premier League football matches. Especially we have evaluated the effectiveness with the Kelly Criterion as a betting strategy. We processed a comprehensive dataset to develop six distinct predictive models, each designed to forecast the probability of three possible match outcomes: a home win, a draw, or an away win. These models included various advanced machine learning techniques, and their performance was enhanced through hyperparameter tuning employing both grid search and randomized search methods. Our methodology involved simulating an entire Premier League season, applying the Kelly Criterion to manage and allocate bets based on the probabilistic outputs of our models. The objective was to evaluate the potential returns on investment from each model. Among the models tested, the neural network emerged as the most successful, yielding a return of 35.48 times the initial bankroll. Other models demonstrated varied levels of success, illustrating the diverse potential of machine learning applications in sports betting. The results of our simulations suggest that while machine learning can indeed be a powerful tool in sports betting, its efficiency depends on several factors. The quality of the input data and the level of sophistication of feature engineering might influence model performance. Our study highlights that further enhancements, such as incorporating more information and/or variables in our data and exploring alternative approaches to feature engineering, could improve predictive accuracy. Additionally, domain knowledge remains a critical component, suggesting that a hybrid approach combining data-driven techniques with expert insight may yield the most robust long-term betting strategies. In conclusion, while our findings are promising, indicating that the application of machine learning and the Kelly Criterion can be profitable, they also underscore the need for ongoing refinement and the integration of comprehensive domain knowledge to fully capitalize on this approach in the competitive sports betting market.}}, author = {{Båth Viderström, Oscar and Sjöberg, Viktor}}, language = {{eng}}, note = {{Student Paper}}, title = {{Beating the Bookies}}, year = {{2024}}, }