A carregar...
Publicação
Machine Learning for Sports Betting on a Betting Exchange: Historical Odds Backtesting
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 22.18 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
The growth of online betting platforms and the increased availability of digitized sports data,
coupled with advancements in technology, have transformed the development of betting approaches, shifting from expert knowledge and statistical analysis to the application of machine
learning techniques for the development of more data-driven, automated prediction models. While many studies have successfully developed profitable models using conventional
bookmakers, these platforms do not offer an ideal environment for bettors, as they impose
restrictions and limitations on winnings. In contrast, Betting Exchanges, such as Betfair, follow a
“winners-welcome” policy, charging only a commission on winnings. The free market nature
of Betting Exchanges leads to more accurate odds that better reflect the true probability of
an event. This creates a challenge for prediction models, as it is harder to develop a strategy
that consistently beats the commission rate. This thesis focuses on predicting whether the
odds for a particular football match will be above or below the closing odds, with the objective
of developing a profitable betting strategy within a Betting Exchange. The research utilizes
historical odds data from Betfair and applies various machine learning techniques — including
Logistic Regression, Decision Trees, Random Forests, HistGradientBoosting, Multi-layer Perceptron (MLP), Transformers, and Long Short-Term Memory (LSTM) models — to analyze the
patterns in odds movements leading up to each match. The best-performing model yielded a
return of 216.45% on the initial bankroll over a span of 25 months, with 313,450 bets placed,
all while accounting for a 2% commission rate. These results demonstrate that developing a
profitable predictive model within the Betting Exchange market is indeed feasible, even in
a “winners-welcome” environment. This thesis not only validates the potential for machine
learning-based sports betting models but also sets the stage for future research. This work can
be used to develop a betting strategy that thrives in real-time in the dynamic and competitive
Betting Exchange environment.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Sports Betting Betting Exchange Odd Machine Learning Deep Learning