A Comparative Study of Motor Imagery (MI) Detection in Electroencephalogram (EEG) Signals Using Different Classification Algorithms

With the advancement of machine learning algorithms, the electroencephalogram (EEG) signal can be highly useful to identify the motor neuron activity (also known as motor imagery, MI event). Depending on the problem and experimental dataset/protocol, researchers use different classification algorithms to classify motor imagery events. Therefore, the selection of classification algorithms for accurate MI detection is significantly important. This paper attempts to provide a brief comparison of the performance of different classification algorithms for MI detection in EEG data. At first, the EEG data of 30 random subjects were collected from a publicly available dataset. Then a combination of time domain and frequency domain features were extracted from the EEG data. Next, the correlated feature selection (CFS) algorithm was used to rank the features based on the correlation. To further reduce the feature set, forward feature selection (FFS) algorithm that was utilized to find out the significant features. For classification, support vector machine (SVM), logistic regression (LR), naïve bayes (NB), and k-nearest neighbor (KNN) were investigated with leave one subject out cross-validation scheme. Finally, the performance of all the classifiers was evaluated using the f1-score. The SVM classifier achieved an average f1-score of 70.26% from 30 subjects' trials. The results suggest that the SVM classifier may perform better than other classifiers. Additionally, the linear kernel of SVM helped to reduce the computational time and complexity of the model. It is expected that the SVM classifier may provide better performance in a wider range of populations.