Comparative Analysis of Methods for Classifying Electroencephalography Signals for a Robotic Brain-Computer Interface

Problem statement: The implementation of a brain-computer interface for robotic systems based on motor activity is carried out by classifying brain activity signals. Electroencephalography (EEG) is the most commonly used method to measure brain activity due to its mobility and high temporal resolution. Currently, extracting meaningful features from a large volume of EEG data in the presence of noise, followed by signal classification, is not a solved problem. Purpose of research: To determine the accuracy of various classification methods for obtaining information about motor activity from EEG data. Method: The performance of classification methods is assessed using the BCI competition IV-dataset 1, which contains EEG data obtained from 9 people performing motor activity tasks. Pre-processing of EEG data consists of finding the spectrum of the EEG signal using the Fast Fourier Transform (FFT) algorithm. The classification methods are implemented in Python. The libraries numpy, tensorflow, keras, sklearn, PIL are used in processing EEG data. Results: The performance of such signal classification methods as Support Vector Machine, k-Nearest Neighbors method, Linear Discriminatory Analysis, Decision Tree Classifier, Convolutional Neural Networks. Practical significance: The information about the performance of various approaches to the classification of brain activity signals makes it possible to create effective systems of brain-computer interfaces for robotic systems.