DETECTİON OF HAND MOVEMENTS BY ANALYZING EEG SIGNALS USING CNN

In this paper a method for detecting six specific hand movement events in a 32-component brain EEG signal by using an ensemble of convolutional neural networks (CN) as a multiclass classifier is considered. The paper proposes and empirically evaluates several options for the architecture of convolutional neural networks, as well as an ensemble that combines the proposed options for the architecture of convolutional neural networks, using a blending algorithm and a final classifier based on logistic regression. The advantages of the chosen classification method for the problem being solved are shown. The results obtained make it possible to say that the use of a classifier in the form of an ensemble of several models of convolutional neural networks allows one to effectively identify characteristic features in the initial EEG signals, and at the output of the classifier to obtain the probabilities that the input signal belongs to one of the given classes of hand movements. The use of the blending algorithm makes it possible to obtain optimal classification results by integrating the best estimates of several models, which individually on the entire test set may give a non-optimal result.