Research Information System
30th Signal Processing and Communications Applications Conference, SIU 2022, Safranbolu, Turkey, 15 - 18 May 2022, (Full Text)
Motor imaginary (MI) task classification based on electroencephalogram (EEG) signals is among the most common brain-computer interface (BCI) studies. One of the most widely used open-access datasets for this purpose is BCI Competition IV Dataset-IIa. In this data set, there are EEG data recorded during MI movements of the left hand, right hand, foot and tongue. In this study, four MI tasks were tried to be differentiated with the classifiers that are frequently used in the literature utilizing these 22-channel EEG data from a total of nine subjects. In addition, the effect of selecting statistically significant features from the features extracted on the dataset was investigated. These feature sets were differentiated using 11 different classification algorithms and 5-fold cross-validation. Each algorithm was tested 10 times to analyzed the repeatability of the results. As a result, classifier performances of %44.38 were obtained in the Ensemble classification Subspace Discriminant algorithm using all time-domain EEG features and %44.00 in the Linear Discriminant Analysis algorithm using only the features selected by ANOVA. Although the highest classifier performance seems to have decreased, it was observed that the feature selection process with ANOVA increased the performance in 6 classifiers, did not change the performance in 1 classifier, and decreased the performance in 4 classifiers. Accordingly, it was concluded that the method of selecting statistically significant features generally increased the classifier performance, but it was difficult to reach a general decision.