A Semi-Supervised Learning Algorithm for Brain-Computer Interface Based on Combining Features

A brain-computer interface (BCI) is a communication system that doesnpsilat depend on brain´s normal output pathways of peripheral nerves and muscles. In this paper, a semi-supervised learning algorithm for BCI based on features combining is proposed aiming at reducing the training process. In order to obtain more stable and effective classification information, two kinds of features are extracted by supervised and unsupervised extractions respectively and two corresponding classifiers are trained. During the learning process of the final classifier, the initial labeled set is enlarged iteratively by unlabeled data (with their predicted labels) whose two labels predicted by both classifiers are same, freeing of setting the threshold of confidence. The features supervised-extracted and both classifiers are updated each iteration for the purpose of absorbing unlabeled data information. At last, the applying on data set I of BCI Competition 2005 demonstrates the validity of our proposed algorithm.