Classification of Brain Activity Patterns Using Machine Learning Based on EEG Data

The article is devoted to the classification of brain activity patterns using machine learning based on EEG data. The aim of the study is to compare the results of machine learning algorithms depending on the chosen strategy for preparing EEG data and extracting features for the training sample with the example of the identifying fists motor activity task. A brief review of scientific works concerning identification of motor and imaginary activity according to EEG data is presented. An overview of the software for viewing and processing EEG records is also given. The key peculiarities of preliminary preparation of the EEG data are analyzed including removal of artifacts and choice of the channels. The approaches to the selection of a set of features for extraction are analyzed. Description of machine learning tools used to classify patterns of brain activity based on the EEG data is given. During the experiment, the following machine learning methods were studied to classify the EEG data: nearest neighbor method, support vector machine, artificial neural network. Pre-processing of EEG data was performed with use of EDFbrowser. To remove artifacts, a Butterworth banpass filter was used. To extract features, Python software libraries for data processing and analysis were used. WEKA-3-8-4 library of machine learning algorithms was used for training classification models.