A Multiclass EEG Signal Classification Model Using Channel Interaction Maximization and Multivariate Empirical Mode Decomposition

Abstract

Brain-Computer Interface (BCI) is an emerging technology that facilitates a pathway between the human brain and external devices. Electroencephalography (EEG) data are mainly employed in BCI systems to reflect the underlying mechanism of different neural activities associated with various limb motions or Motor Imagery (MI) activities. Multichannel EEG signal processing generally results in high-dimensional features, which increases BCI’s overall computational and temporal complexity. We introduce a channel selection methodology using the mutual information-based three-way interaction scheme to reduce this burden due to many channels. Our approach initializes a set of three candidate solutions for a given MI classification task and subsequently determines a highly significant EEG channel set. It effectively balances relevance and redundancy levels in the final channel subset during the selection and rejection of a newly selected channel. The proposed scheme is evaluated on the BCI Competition IV-2008 dataset with four MI classes (left hand, right hand, tongue, and feet) and twenty-two channels. The performance of our scheme is compared with three recently published state-of-the-art methods. The proposed approach realized an average of 86.66% classification accuracy using only nine channels on the data of nine participants. The comparative study shows that our approach realized better performance in terms of higher classification accuracy and channel reduction rate than all three baseline models. The results are promising for the online BCI paradigm that requires low complexity while conducting multiple sessions of BCI experiments for a larger group of participants. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.