Research Paper AAMAS 2020, May 9–13, Auckland, New Zealand Optimising Game Tactics for Football Ryan Beal Georgios Chalkiadakis
[email protected] [email protected] University of Southampton, UK Technical University of Crete, Greece Timothy J. Norman Sarvapali D. Ramchurn
[email protected] [email protected] University of Southampton, UK University of Southampton, UK ABSTRACT Prior multi-agents research for football has focused more on the In this paper we present a novel approach to optimise tactical contribution of individual agents within a team [3, 8]. However, to and strategic decision making in football (soccer). We model the date, there is no formal model for the tactical decisions and actions game of football as a multi-stage game which is made up from a to improve a team’s probability of winning. There are a number Bayesian game to model the pre-match decisions and a stochastic of tactical decisions that are made both pre-match and during the game to model the in-match state transitions and decisions. Using match that are often just made through subjective opinions and this formulation, we propose a method to predict the probability “gut feelings”. of game outcomes and the payoffs of team actions. Building upon Against this background, we propose a formal model for the this, we develop algorithms to optimise team formation and in- game of football and the tactical decisions that are made in the game tactics with different objectives. Empirical evaluation of our game. We model the game as a 2-step game that is made up of a approach on real-world datasets from 760 matches shows that by Bayesian game to represent the pre-match tactical decisions that using optimised tactics from our Bayesian and stochastic games, we are made due to the incomplete information regarding the tactical increase a team chances of winning by 16.1% and 3.4% respectively.