Research on Football Team Ability Based on Random Forest and Ranking Method

Abstract

The outcome of a football match covers many skills between individual ability and team strategy. This paper is an in-depth analysis of football team strategy. Firstly, the passing network between players is established, and the passing network is used to determine the formation of players. Next, it is determined that the performance indicators that reflect successful team cooperation, including the diversity of game types and the coordination of players, and then determine other team-level processes. Finally, create a model according to the determined performance indicators and team process, and use the model to analyze the structure, configuration, and dynamic characteristics of the team operation. In the experimental stage, through the observation and analysis of team cooperation mode, this paper puts forward the optimization structure strategy to achieve greater efficiency, and helps the coach improve the success rate of the team through network analysis. In this problem, according to all the matches of the four FIFA world cups from 2002 to 2014, we compared three different modeling methods for predicting the score of football matches: the Poisson regression model, random forest and the ranking method. The first two methods are based on the covariate information of the team, while the latter method estimates sufficient capability parameters, which can best reflect the strength of the current team. In this comparison, the prediction method with the best performance on the training data finally becomes the ranking method and random forest. By combining the random forest with the team ability parameters of the ranking method as additional covariates, the prediction ability can be greatly improved.