A New machine learning approach for in-game behavior analysis of football players based on player statistics

Abstract

This research endeavors to leverage technological innovations to enhance player evaluation and recruitment processes within football clubs. The study commenced with the acquisition of player attributes and ratings from the FIFA24 dataset, supplemented by additional statistics sourced from various reliable sources. Player performance ratings were obtained from Sofascore ratings, aim to create a unified dataset. Through data preprocessing and attribute selection the dataset was refined to include only relevant variables for analysis. The research proceeded with the development of predictive models for player performance ratings, utilizing advanced machine learning techniques. A subset of the dataset was designated for training these models, and various algorithms were explored and evaluated to identify the optimal ensemble model, resulting in voting method ensemble model predicting player performance ratings. Furthermore, player clustering based on player characteristics ratings was undertaken using the K-means clustering algorithm. The optimal number of clusters was determined through the elbow method, and distinct player clusters were identified based on their playing characteristics. A random forest model was trained with accuracy of 96% to predict player clusters, enabling the categorization of players into distinct playing styles. This research effort led to the development of a comprehensive system capable of predicting player performance ratings and clustering players based on their playing styles. This system offers valuable insights to football clubs, aiding in informed decision-making processes related to player recruitment and team composition. By using data analytics and artificial intelligence, football clubs can gain a competitive advantage in identifying talents suitable for them, ultimately enhancing the performance and success of their teams on the field.