Soccer simulation is an effort to motivate researchers to perform artificial and robotic intelligence investigations in a multi-agent system framework. In this paper, we propose a game theoric-based data mining approach to help the coach agent select the best strategy for each soccer player agent in order to gain the most probable payoffs.These payoffs are calculated both static and dynamic i.e. are taken from experience results that are stored in a knowledge-base or is learned knowledge during the game. In this work we have confined ourselves to a model in which opponent strategy remains static. We take advantage of a learning algorithm with a polynomial time complexity in the number of states of the opponent strategy modeled by deterministic finite automata.