Abstract

This paper explores the problem of EEG channel evaluation and selection in P300-based brain-computer interface (BCI). A new channel importance evaluating method by rough set was proposed and applied to P300 dataset collected under a P300 Chinese typewriter paradigm. Channels in midline and posterior brain achieved higher significance, which accords with classical channel selection in P300 BCI studies. Then channel sets selected by three evaluation methods based on rough set, F-score and classification accuracy of Fisher's linear discriminant analysis (FLDA) were compared using both support vector machine (SVM) and FLDA classifiers. The highest accuracy was obtained by rough set method and channel sets selected by rough set achieved slightly better performance overall in P300 detection.

Paper Details