Brain-Computer Interface of Motor Imagery Using ICA and Recurrent Neural Networks

Abstract

Brain-Computer Interface (BCI) is a device that can connect brain commands without the need for movement, gesture, or voice. Usually, BCI uses the Electroencephalogram (EEG) signal as an intermediate device. EEG signals need to be extracted into waves that represent the action in mind. In this study used Wavelet transformation to obtain the imagery motor component from the EEG signal. However, the problem also arises in the considerable channel redundancy in EEG signal recording. Therefore, it requires a signal reduction process. This paper proposed the problem using Independent Component Analysis (ICA). Then ICA components are features of Recurrent Neural Networks (RNN) to classify BCI information into four classes. The experimental results showed that using ICA improved accuracy by up to 99.06%, compared to Wavelet and RNN only, which is only 94.06%. We examined three optimization models, particularly Adam, AdaDelta, and AdaGrad. However, two optimization models provided the best recognition capabilities, i.e., AdaDelta, and AdaGrad.