Comparison of Motor Imagery EEG Classification using Feedforward and Convolutional Neural Network

Brain-computer interface (BCI) is widely used in several clinical applications. Motor imagery-based BCI can help patients who have lost their motor functions in communication and rehabilitation. To develop such BCI applications, the accurate classification of motor-imagery based electroencephalography (EEG) is crucial. By processing a publicly available EEG dataset, we obtained information that can be used to train neural networks and efficiently classify activities performed by volunteers. In this paper we used several data pre-processing methods and examined how they affect the classification performance of a feedforward neural network. As the results were not satisfactory with the feedforward network, the data prepared with the best pre-processing method were also used to train a convolutional neural network (CNN). We achieved an accuracy of 91.27% in classifying fists and feet closing activities using data from ten volunteers.