Quantifying football player value: a novel combined rating and ML ensemble approach for predicting top players

Abstract In football (soccer), players are considered one of the most important elements of a club’s sporting and financial success. Therefore, it is extremely important to evaluate the (sporting) value and future potential of a player. Usually, scouts and other decision-makers are responsible for this assessment, which is generally based on laborious observations and subjective experience. In recent years, advances in the data-driven evaluation of professional football players – including the use of Machine Learning (ML) – have led to a transformation that has the potential to make this process more objective and resource-efficient. Based on this, we present a novel and interpretable combined rating for evaluating the player strength of professional football players, considering the real performance of the players from a team and individual perspective. To provide further transparency, our combined rating is assessed with regard to the quality criteria of objectivity, reliability, and validity. Thus, our proposed rating achieved an RMSE of 0.486 on an out-of-sample retrodiciton test, outperforming other comparable metrics and demonstrating that it is a strong predictor of player performance. In addition, based on our proposed rating, we predict the future development potential of players, presenting a ML ensemble that estimates the likelihood of a player becoming a top player or not. Our approach achieved a balanced accuracy of 83.90%, representing the first of its kind and setting a benchmark. In addition, the top 30 talents for the 2024/25 season are presented to illustrate the effectiveness of our approach. Finally, based on our findings, optimization approaches are discussed on which future research can build on.