Abstract: Game is a kind of behavior based on information and experience to maximize the interests. The brain research on the game has a long history. The methods of research are various, but they all focus on the results of the principle. Almost no one is involved in the process of brain activity. Therefore, this paper designed a “rock-paper-scissors" experiment paradigm, which has five stages of spontaneous, evaluation, decision making, feedback and rest. We acquired EEG of 17 subjects and extracted time-frequency characteristics of brain cognitive process. Through the analysis of brain activity, a predicting method is proposed based on the characteristics of game decision optimization. And SVM has been used to predict the decisions and achieved a recognition rate of 80%. It shows that the optimal feature proposed in this paper could reveal the characteristics of the cognitive process of the game.