Using Self-organizing Map for Mental Tasks Classification in Brain-Computer Interface

Abstract

One problem in Brain-Computer Interface (BCI) is the requirement of online training of classifiers, since EEG patterns vary greatly at two separate time with long period. In this paper, the use of Self-Organizing Map (SOM) as an adaptive classifier for mental tasks classification was proposed. As for SOM, there are two cases about the labeling of map units, which correspond to semi-supervised and unsupervised algorithm respectively. In one case, the map units are labeled according to the labels of training patterns. In the other case, the map structure information, e.g., the U-matrix, is used to cluster map units. The ability of SOM to recognize mental task was analyzed for both cases. The organized SOM is tested on testing patterns. The averaged classification accuracy of 96.2% and 90.8% across 10 task pairs was obtained for both cases respectively. This result indicates the feasibility of online training of SOM for mental tasks classification.