Análise de concordância sobre previsões de resultado de partidas de futebol

Abstract

We live in a time of high turnover of money in the sports betting market, and with this the recent interest in using machine learning techniques for predicting sports outcomes has grown. This paper, using a database of Premier League games, evaluated the use of a set of machine learning algorithms to predict sports results and assess the e↵ectiveness of di↵erent levels of agreement of the predictions of the di↵erent models used. The results of the predictions after analyzing the agreement between the predictions of the models and the di culty in predicting draws were highlighted in the paper. As a more significant result, we had an accuracy rate of 83% after excluding the games that did not have 100% agreement in the answers between the trained models. Besides this, future works were suggested, such as the inclusion of other variables and characteristics of the games, the generalization of the tests to other leagues, and a new approach to the problem of predictions in tie cases.