A Novel BCI System Based on Hybrid Features for Classifying Motor Imagery Tasks

Brain-Computer Interface (BCI) is a way to control external devices based on Electroencephalography (EEG) signals. One of the most critical problems facing BCI is realizing high Classification Accuracy (CA) for Motor Imagery (MI) mental tasks. A novel study is proposed which aims to achieve a reliable CA. In this study, three sets of features were extracted in time, time-frequency, and the time and time-frequency domains. Several Support Vector Machine (SVM) classifiers were constructed with different electrode sets to determine the channels which improve the CA. The publicly available dataset BCI competition III datasets Iva was used in this study. The results showed that the proposed method is one among the few, which focuses on achieving higher classification accuracy depending on features from different domains. The highest mean CA of 91.72% was achieved using the hybrid feature set extracted from the time and time-frequency domains. The mean CA of the proposed outperformed other several recent related works. Therefore, the proposed can be used successfully to control wheelchairs and rehabilitation therapies.