EEG-Based Machine Learning: Theory and Applications

Abstract

Electroencephalography is a widely used clinical and research method to record and monitor the brain's electrical activity - the electroencephalogram (EEG). Machine learning algorithms have been developed to extract information from the EEG to help in the diagnosis of several disorders (e.g., epilepsy, Alzheimer's disease, and schizophrenia) and to identify various brain states. Despite the elegant and generally easy-to-use nature of machine learning algorithms in neuroscience, they can produce inaccurate and even false results when implemented incorrectly. In this chapter, we outline the general methodology for EEG-based machine learning, pattern recognition, and classification. First, a description of feature extraction from various domains is presented. This is followed by an overview of supervised and unsupervised feature-reduction methods. We then focus on classification algorithms, performance evaluation, and methods to prevent overfitting. Finally, we discuss two applications of EEG-based machine learning: brain-computer interface (BCI) and detection and prediction of microsleeps.