Sanghoon Shin, Moohong Min, Journal of Internet Computing and Services, Vol. 27, No. 1, pp. 49-65, Feb. 2026
10.7472/jksii.2026.27.1.49, Full Text:  HTML
Keywords: Machine Learning, Soccer, Playing Styles, clustering, Classification Model

Abstract

This study developed classification models to quantitatively predict soccer players' playing styles using machine learning based on match log data. Match log data for each player provided by FBref, a soccer statistics website, were collected and used in the study. Cluster analysis was conducted on the collected data to classify player types by playing style for each soccer position, and a soft clustering method, Fuzzy C-Means (FCM), was applied to reflect the diverse playing styles of the players. Based on the derived results of the cluster analysis, playing style types were labeled, Random Forest, Logistic Regression, Gradient Boosting, XGBoost, and LightGBM models were trained employing a multi-label classification method, and the predictive performance of each model was compared by position. As a result, the XGBoost model demonstrated the best predictive performance, achieving the Macro F1-Score of over 0.93. In addition, using SHAP (SHapley Additive exPlanations) plots, feature importance and the correlations between features and playing styles were analyzed to identify the key factors determining playing styles and to enhance the interpretability of the model. This model contributes to quantitatively explaining players' playing styles, which have relied on subjective evaluations, and is expected to provide useful information in decision-making processes such as player scouting and tactical planning.

Statistics
Cite this article