Multi-segment Majority Voting Decision Fusion for MI EEG Brain-Computer Interfacing

Brain-computer interfaces (BCIs) based on the electroencephalogram (EEG) generated during motor imagery (MI) have potential to be used in brain-controlled prosthetics, neurorehabilitation and gaming. Many MI EEG classification systems segment into windows for classification. However, a comprehensive analysis of decision fusion segmented data, within context different classifiers, has not been carried out. This study presents multi-segment majority voting (MSMV) approach which an trial is using overlapping windows. Segments are labelled final label derived through voting, common spatial pattern (CSP) features. The impact MSMV accuracy six classifiers was investigated. effects window size overlap were analysed. Results five subsets channels, channel static also proposed. BCI Competition III dataset IVa used. found significantly improve linear discriminant (LDA), support vector machine (SVM), naïve-Bayes (NB) random forest (RF) classifiers. improved by 5.02%, 4.41%, 1.25% 3.62% SVM, LDA, NB RF respectively. indicated importance central-parietal central-frontal electrode regions performance could considered future studies, particularly online that deal with buffered data.