Abstract
Brain computer interface (BCI) for controlling outer machine by brain signal has been getting an attractive topic recently. It is necessary to classify observation data to conduct a control command when outer machine is controlled by brain signal, and we use clustering method to identify it generally. However, the accuracy of identification class is not high if the amount of observation data is small. In addition, the outer machine will not follow to a change of environment dynamics when the identification class was discriminated by past data. In this paper, we propose a permutation data method which is called PDSM (Permutation Data Structure Method). PDSM determines the permutation of electroencephalographic signal data for improving accuracy of identification class when the observation data is learned consecutively. In PDSM, we select sampling data from observed data for testing learning, and change data permutation to improve accuracy of identification class. We propose 15 kinds of PDSMs by the viewpoint of strategic methods which distinguish data pattern and data structure which the changes data order and the class order. We discuss usefulness of PDSM by applying numerical sample data to PDSM and estimating accuracy of identification class.
