Channel Selection and Classification of Electroencephalography based Motor Imagery Signals using Machine Learning

Brain Computer Interface (BCI) is a topic of research for many years. It is beneficial for people with limited mobility, in addition BCI is also applied in other fields such as entertainment, security, education. Electroencephalography (EEG) is the most utilized brain signals in BCI systems. The accurate classification of EEG signals is highly desirable for BCI applications. In this study motor imagery (MI) EEG signals are classified into left- and right-hand movements. Publicly available dataset from PhysioNet project is utilized in this study. Four kinds of features including band power, approximate entropy, statistical features, and wavelet features were extracted from the data. These features were used individually as well in different combinations to test their reliability in classifying the MI tasks. Three different classifiers, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Convolutional Neural Network (CNN) were used to classify the data. The capability of CNN to classify the raw signal instead of extracted features from the data is also investigated. Additionally, the performance of the CNN model with the reduced number of EEG channels is tested. The best result was obtained using CNN as a classifier with statistical features, with a classification accuracy of 94.28%. The effect of different combinations of features and the reduced number of EEG channels are discussed in the paper.