Classification of electroencephalographic records during motor movement imagery

The methods for communication between brain and computer represent an alternative solution for management of a computer. Such solutions are suitable for persons who are physically unable to use typical mechanical input devices such as keyboard and mouse. In the scope of diploma thesis, we developed a method for moving the cursor on a computer screen on the basis of the electroencephalographic (EEG) records recorded during motor movement imagery, and during actual closing of the left and right hand. We used the EEG records of the publicly available, annotated, EEG Motor Movement Imagery DataSet (EEGMMI DS) database, which is freely available on the Physionet pages. The developed method incorporates digital signal processing procedures, feature extraction procedures, machine learning approaches, and classification of the segments of the records with the aim to classify between the imagined or actual actions of the left and right hand. A very important result of the research gave the use of the sequential forward feature selection procedure. The highest classification accuracies were achieved using the selected features obtained from those signals which correspond to the electrodes F7, F8, FT7 and FT8, and not from the signals which correspond to the traditionally used electrodes C3 and C4. In the research, we achieved the average classification accuracy of 66.78 % for all records of the EEGMMI DS database, while the highest classification accuracy for the records of an individual subject was 97.62 %