Prediction of football match outcomes from passing network structure

• Ano de defesa: 2024
• Autor(a) principal: Xavier, Felipe Jordão
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Aprendizado de máquina; Ciência das redes; Complex networks; Football; Futebol; Machine learning; Network science; Passing networks; Redes complexas; Redes de passe; Soccer
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-28082024-162138/

The majority of methods for predicting match outcomes in football have been developed using generic historical data, such as shots on goal, number of fouls and number of cards, with little regard to the organisation of the teams. With the recent evolution of tracking technology, the collection of precise event data for a match has been made possible, allowing for the construction of passing networks detailing the interaction between players. In this work, we develop a method to predict outcomes of matches from the passing network structure of the teams. For this end, we build the passing networks from event data and train different machine learning algorithms to predict match outcomes from the metrics of such networks. We then rank those metrics based on their impact on the predictive model output. Our method achieves a mean accuracy of 58.5%, compared to a baseline accuracy of 44.8%, on a data set containing 1941 matches from five of the most influential leagues in Europe, the 2018 World Cup and the 2016 Euro Cup. The most important metrics for successful teams are the largest eigenvalue of the adjacency matrix, degree distribution statistics and the average shortest path length.