Movement imagery classification based on subband BSS

In the EEG signals, information is contained in a narrow frequency band. How does one selects the number of subbands either overlapping or non-overlapping and its bandwidth in context with the EEG signals is a key problem in the subband BSS. The authors propose a novel algorithmic approach to estimate the number of subbands and its bandwidth and applied to movement imagery classification. To ensure the perfect classification between the left and right imagery data, the authors propose a novel class performance index (CPI) to select the final separating matrices over a four unique pair under the supervised learning approach.