Prediction of basketball match results by machine learning methods

Basketball is one of the world's most watched sporting events. Because of this prominence, the use of information technology in basketball is frequent. With the advancement of data science and data storage systems, it is now very easy to store parameters such as match statistics and player features. Predicting match results is critical in terms of giving people pre-competition information. While studies are available about prediction of basketball match results for American leagues, works are rare for European leagues. In this study, prediction of Turkey Men's Basketball Super League match results is studied using machine learning methods. Accordingly, the matches played during the seasons 2013-2014 to 2017-2018 were used as 5 separate data sets and analyzed along with the Four Factor and DefenseOfense models. The data were evaluated kNN, Logistic Regression, Multilayer Neural Network, Naive Bayes, j48 and Voting machine learning methods. As a result, a predictive performance of 96.55 percent has been achieved. Additionally, the matches played during the seasons of EuroLeague 2012-2013 and 2016-2017 were used as 5 separate data sets and analyzed along with the Four Factor and DefenseOfense models. The data were evaluated by kNN, Logistic Regression, Multilayer Neural Network, Naive Bayes, j48 and Voting methods and 98.90% prediction success was achieved. The results show that the combination of the Four Factor and DefenseOfense models gives better accuracy rates for both leagues.