Machine Learning Value Prediction based on a Soccer Player's Performance on the Field

Football is the most popular sport worldwide. With the development of the global economy, it has evolved beyond a competitive activity into a rapidly expanding market. Especially for frequent market transfers, the value of players has become a hot topic of concern to everyone. This paper employs machine learning to analyze the huge amount of player performance data on the field to predict the value of players. The data in this paper comes from a professional football data website, T Football Website. It mainly crawls various data including offense, organization, defense, etc. of players in the five major leagues in the 23/24 season. Feature selection is carried out by analyzing the correlation of the features, and then the target variable is logarithmically transformed to reduce the scale and alleviate skewness and other preprocessing. In terms of the model, this paper uses three different models, XGBoost, GBDT, and RF, to predict the player's value. Model performance is evaluated by comparing the predicted value with the true value scatter plot, as well as indicators such as R², MAE, and RMSE. The player's value is evaluated through 54 player instances, and the cause of the error is analyzed. The results indicate that the three different models are all feasible for this prediction to a certain extent, but due to factors such as the international reputation of the players, the rise and fall of their values during the growth and decline periods, etc., the models have a certain degree of error.