Brain-Computer Interface: Common Tensor Discriminant Analysis classifier evaluation

The performance of the Common Tensor Discriminant Analysis method for Brain-Computer Interface EEG pattern classification is compared with three other classifiers. The classifiers are designed with the aim to distinguish EEG patterns appearing as a result of performance of several mental tasks. Classifier comparison has yielded quite similar results as regards our experimental imagery movement data set as well as for BCI Competition IV data set. The Bayesian and Multiclass Common Spatial Patterns classifiers, which use solely interchannel covariance as input, are shown to be comparable in performance, while lagging behind the Multiclass Common Spatial Patterns classifier and the Common Tensor Discriminant Analysis classifier, that is classifiers which additionally account for EEG frequency structure. It is shown that the Common Tensor Discriminant Analysis classifier and the Multiclass Common Spatial Patterns classifier provide significantly better classification than other two methods but at a higher computational cost.