Abordagem baseada em gradient boosting para previsão de resultados em corridas de galgos

Abstract

Greyhound racing, a traditional sport in the United Kingdom, attracts both enthusiasts and bettors seeking profit in this field. In 2020, this market generated over £230 million in on-site and off-site betting. While many individuals rely solely on intuition, more experienced bettors employ techniques to analyze greyhounds’ historical performance in an attempt to predict the winner, a task that can be challenging for beginners due to the complexity of the data involved. This study investigated the application of machine learning techniques for predicting outcomes in greyhound races, by developing a predictive model capable of estimating race times using the CatBoost regression algorithm. This allowed for the projection of race rankings and the identification of potential betting strategies within this market. The analysis involved the construction of a data preparation pipeline inspired by relevant works in the current literature, as well as the use of hyperparameter optimization techniques during model training. To ensure the model’s ability to generalize to unseen data, prediction accuracy was evaluated through k-fold cross-validation. This study contributes to the advancement of knowledge in the development of predictive models in machine learning applied to heterogeneous datasets, and to the identification of sports betting strategies in greyhound racing.