Statistical Learning for the Modeling of Soccer Matches

For roughly the past 30 years, statistical and machine learning techniques have been increasingly used to model and predict soccer matches. There are different ways to define a meaningful response variable. Depending on the choice of this response variable, different approaches of statistical or machine learning are suitable for modeling and predicting soccer matches. While in the beginning mainly classic regression methods were used, machine learning methods like extreme gradient boosting or random forests have been applied more frequently in recent years. These methods were able to improve the predictive performance for new matches in the future, but they are also more complex and (sometimes considerably) more difficult to interpret.