Abstract
This study offers a comprehensive analysis of both classic and advanced features of football players, aiming to enhance player evaluation and feature prediction. We applied a three-step methodology: first, hierarchical clustering was used to reveal the group structure of features in each position. Second, representative features were identified within clusters leveraging the correlation matrix of features, reducing the dimensionality of the dataset while retaining critical information. Third, several regression models were employed to predict other player features using the representatives. This approach was applied across the distinct positions of goalkeeper, defender, midfielder, and forward. Bootstrap resampling confirms the robustness of the results obtained, revealing consistent clusters against random data variations. The findings indicate that representative features effectively encapsulate the entire feature space for each position, allowing other features to be predicted accurately with minimal errors. This study contributes to football analytics by providing a robust method for feature selection and prediction, ultimately improving the accuracy and efficiency of player performance analysis.
', './data/cmcma/coversheet/cover_en.jpg'))