Data-Driven Frequency Bands Selection in EEG-Based Brain-Computer Interface

In this paper, we propose a novel method of frequency bands selection based on the analysis of a channel-frequency map, which we call 'channel-frequency map'. The spatial filtering, feature extraction, and classification processes are operated in each frequency band in parallel. We determine a class label for an input EEG based on the outputs from the multi-streams with a two-step decision strategy at the end. From our experiments on a public dataset of BCI Competition IV (2008) II-a that includes four motor imagery tasks from 9 subjects, the proposed algorithm outperformed the Common Spatial Pattern (CSP) algorithm and a filter bank CSP algorithm on average in terms of a session-to-session transfer rate using one session for training and the other session for test. A considerable increase of classification accuracy has been achieved for certain subjects. We also would like to note that the proposed data-driven frequency bands selection method is applicable to other single-trial EEG classification that is based on modulations of brain rhythms.