ABSTRACT
Independent Component Analysis (ICA) is often used at the signal preprocessing stage in EEG analysis for its ability to filter out artifacts from the signal. The benefits of using ICA are the most apparent when multi-channel signal is recorded. The question is, however, what kind of benefits (if any) can be obtained when ICA is applied for a few channel recording. We addressed this question in this paper by setting up the hypothesis that even in the case of only three channels, ICA can rearrange the sources to new mixtures in such a way that the true brain sources will be enhanced in some components, and the artifacts will be enhanced in others. To verify our hypothesis we applied three popular ICA algorithms to preprocess data from a benchmark file (motor imagery file from the II BCI Competition). Our results, presented in terms of classification precision, show that all ICA algorithms enhanced the signal to noise ratio for components correlating with signals recorded over C3 and C4 channels (the classification precision was higher in their case) and lessened the signal to noise ratio for components correlating with signals recorded over Cz channels.