An XGBoost-lasso ensemble modeling approach to football player value assessment

Abstract

Sports analytics is a field that is growing in popularity and application throughout the world. One of the open problems in this field is the valuation of football players. The aim of this study is to establish a football player value assessment model using machine learning techniques to support the transfer decisions of football clubs. The proposed model is mainly based on the intrinsic features of the individual players which are provided in Football Manager simulation game. To do this, based on the individual statistics of 5316 players who are active in 11 different major leagues from Europe and South America, different value assessment models are conducted using advanced supervised learning techniques which include ridge and lasso regressions, random forests and extreme gradient boosting. All the models have been built in R programming language. The performances of the models are compared based on their mean squared errors and their fit to the real world examples. An ensemble model with inflation is proposed as the output.