Research On Game Decision Prediction Method Based On Eeg Signal

Game is a process of making decision, which we try to make it benefit to ourselves, according to the information they have, when competing with each other. In recent years, with the rapid development of the psychology, physiology, neuroscience and the computer science, the state of game cognitive play a very important role in the research.According to the process of the experiment, we divided the game into different stages based on the different activities of the states of the brain. Based on experiment design, we gather the eeg of the 17 participants. The eeg data can’t avoid to the blink, horizontal and vertical eye electricity, electromyography and so, on the trail. So there is high needed to remove the artifact correction. In this paper, we adapt the double ICA to make it. And we evaluate the results in both the time and the space.For the sake of making the processing state of the game cognitive, “stone scissor paper”, more clearly, we traced to the source of the experiment. From both the time and frequency domain, we traced it.In the time domain, in the earlier of the game(0~4000ms), we found the brain had been activated relates to visual information processing, as well as the high level information processing areas of the brain, and the motor cortex is inhibited. Game activities during the second half of the time(4000~5000ms) it is related to the movement of the motor cortex activation. Responsible for visual information processing, as well as the high level information processing areas of the brain are suppressed. From the frequency domain, we analysis spectrum is related to game Alpha, Theta and Gamma in the "stone scissors paper" game activities.On the basis, we extract the stone, scissor and paper’s statistics features of EEG, from time domain, frequency domain and the time-frequency. Lastly, we select features by using feature selection method based on Correlation-based Feature Selection. Finally, use the SVM classifier for classification recognition, with linear kernel function as the SVM kernel function, and the recognition rate reached 80%.