Implementing a fuzzy inference system in a multi-objective EEG channel selection model for imagined speech classification

摘要: It was searched the minimal subset of channels for imagined speech.Channel selection approached as multi-objective to obtain a Pareto front.A fuzzy system inference applied find promising solution from front.Channel had statistically similar performance use all channels.It observed dependence between features and classes speech. One main purposes brain-computer interfaces (BCI) is provide persons an alternative communication channel. This objective firstly focused on handicapped subjects but nowadays its scope has increased healthy persons. Usually, BCIs record brain activity using electroencephalograms (EEG), according four neuro-paradigms (slow cortical potentials, motor imagery, P300 component visual evoked potentials). These analytical paradigms are not intuitive difficult implement. Accordingly, this work researches neuro-paradigm called speech, which refers internal pronunciation words without emitting sounds or doing facial movements. Specifically, present research recognition five Spanish corresponding English "up," "down," "left," "right" "select", with computer cursor could be controlled. We perform offline automatic classification procedure dataset EEG signals 27 subjects. The method implements channel composed two stages; first one obtains front optimization problem dealing error rate number channels; second stage selects single (channel combination) front, applying (FIS). assess method's through combination test set used generate front. Several FIS configurations were explored evaluate if able select combinations that improve or, at least, keep obtained accuracies each subject's data. found configuration, FIS3×3 (three membership functions both input variables: channels), best trade-off rules accuracy (68.18% around 7 channels). Also, compared (70.33%). Results our demonstrate feasibility automatically speech classification. presented outperforms previous works in showed relationship data words.