Currently, there are no uniform recommendations for setting up both recording equipment and software to ensure the uniformity of measurements and increased accuracy of classifications in motor image recognition tasks. However, improving the accuracy and consistency of classifications is the main goal of all research in this field, regardless of the registration equipment and software used. The purpose – development of a unified methodology for configuring the parameters of the brain-computer interfaces to ensure the uniformity of measurements and the maximum possible values of accuracy in the classification of motor images. Recommendations are given on setting up the electroencephalogram recording equipment. To form recommendations on options for optimal localization of electrodes, as well as their number, computational experiments were conducted to calculate the accuracy of classification of motor images when processing signals received from various numbers of electrodes and their localizations. As a result of the experiments, it was found that when the reference electrode is located in the central part of the head, the number of pairs of electrodes to achieve the highest classification accuracy should be from 3 to 5 (respectively, these are 6-10 electrodes located symmetrically relative to the midline of the head). The influence of the frequency range of EEG registration on the classification accuracy was also investigated. As a result of the study, it was found that the maximum amplitude of the beta rhythm oscillations that desynchronize during the preparation and execution of the motor act are in the frequency band 19-31 Hz. However, this range strongly depends on the physiological properties of the subject and can be adjusted. The influence of three machine learning algorithms on the accuracy of classification has been studied. The optimal settings of the classifiers have been formed. The maximum average classification accuracy was 67.37% when using a classifier based on the linear discriminant analysis (LDA) method. The maximum peak value of classification accuracy was 74.90%. When using a third-party database, the average classification accuracy was 73.44%, which shows the correctness of the recommendations formulated in the methodology. The developed technique will be useful to a wide range of researchers and developers of brain-computer interfaces due to the fact that setting up the equipment will take much less time. This will allow us to focus on the main task of research in this area – the creation of a universal classifier architecture that provides increased accuracy. The recommended technique for configuring neurocomplexes in MI recognition will allow you to achieve maximum classification accuracy, regardless of the type of electrodes, the type of registration equipment used and classification software.
Zhuravlev et al. (Mon,) studied this question.