Analysis of soccer playerʼs positioning patterns by Hidden Markov Model

Abstract

    It has been an important issue to develop quantitative method for analyzing team tactics in football games. In this paper, time-series data of the ball and the player positions were stochastically represented by using Hidden Markov Model (HMM). First, the time-series data was classified into two patterns of effective offense and non-effective offense based on the offensive performance after the ball takeover. HMM for each offense pattern was constructed by using machine learning method from the time-series training data. Then, the time-series data was automatically classified into two offense patterns by using HMM classification method. The classification rate was more than 90％ for the training data and more than 80％ in the testing data. Also, the classification test was performed for a part of the time-series data in the interval from the ball takeover to the instant when the ball was carried at a reference pitch line. The classification rate was approximately 70-90％ for the training data and 70-85％ in the testing data in cases where a certain number of states were set up in HMM. Moreover, the occurrence probability of HMM was compared between the effective offense and the non-effective offense. The result showed significant differences of the playerʼs formations between two offense patterns. These results demonstrated that HMM modeling for the time-series data was an effective method to analyze the team performance in football games.