From Pitch to Paycheck: Predicting Premier League Salaries with Player-Grouped Machine Learning Validation

Football players’ salaries are associated with the individual player’s performance as well as the environment in which they play. This study investigates how well public performance and demographic data can be utilized to predict salaries of English Premier League players while allowing for repeated player data. Data utilized in the study included 1,240 salary records of 672 individual players for the three EPL seasons from 2022 to 2025. Euro-denominated salaries were converted to log10 values. Seventeen different numerical predictors and three categorical predictors were used in eight separate regression models. Five-fold group-based cross-validation was employed, preventing the same player from appearing in both training and validation sets. Of the eight models tested, Elastic Net had the highest performance, with a mean R2 of 0.420, a mean absolute error of 0.226 log10 units, and a multiplicative error factor of 1.68. A temporal split wherein the first two seasons were used for training and the 2024-25 season was used as the test produced similar performance levels for Elastic Net, with R2 = 0.435 and an error factor of 1.64. Feature importance analysis indicated that the player’s current club and player age were the strongest predictors, whereas nationality, number of games started, and expected goals scored per 90 minutes also contributed to prediction. The public dataset provided a moderate ability to predict salaries, but the relationships are not causal in nature. Club context and player characteristics were more impactful predictors of salary than any specific metric of match performance.