Football analytics using Bayesian networks: The FutBA model

The results of football matches are difficult to predict due to their high uncertainty. Previous applications of data-driven machine learning approaches had limited performance in this prediction problem. Models that use expert knowledge had relatively higher performance but it is difficult to adapt these models to different cases as they need to be reviewed by experts and analysts based on specific requirements of the new application. This paper proposes a novel Bayesian network model to predict the results of football matches in Turkish football leagues. The Bayesian network model predicts the match results by estimating the attack and defense capability of the teams based on multiple observations about the football match. The structure and parameters of the model is defined based on expert knowledge. The model is able to use statistical data from previous matches and expert knowledge about these factors to generate predictions. The proposed model is evaluated by using data from the Turkish Super League