Machine Learning approach for the classification of EEG signals of multiple imagery tasks

Abstract

Electroencephalogram (EEG) signals can be used to capture the electrical pattern generated on the surface of the human brain. The electrical activity in terms of EEG signals occurs in the brain as a result of any type of emotions, behavior, thoughts that occur in the human brain. These EEG signals can be utilized as an input to the Brain-Computer Interface (BCI) application. The proper feature extraction and classification of the extracted features are a very important step for BCI applications. In this paper, the collection and recording of EEG signals are done corresponding to eight imagery tasks. The extraction of the features is done in the form of absolute band powers, based on the logarithm of the Power Spectral Density (PSD) of the EEG data for each channel. The selected features are used for the classification and analysis of the captured EEG signals corresponding to the eight cognitive imagery tasks such as 'forward', 'backward', 'left', 'right', 'hungry', 'food', 'water', and 'sleep'. The selected features are then used to train the Machine Learning (ML) models of Logistic Regression (LR), and Quadratic Discriminant Analysis (QDA). The classifiers are analyzed and investigated on several evaluation metrics such as the average accuracy of the classifiers, accuracy of each imagery tasks, precision, and recall. The average accuracy achieved by the classifier is 72.5%, and 74.7% for LR, and QDA respectively. © 2020 IEEE.