Review of EEG Signals Classification Using Machine Learning and Deep-Learning Techniques

Electroencephalography (EEG) signals have been widely used for the prognosis and diagnosis of several disorders, such as epilepsy, schizophrenia, Parkinson’s disease etc. EEG signals have been shown to work with machine learning techniques in the literature. However, they require manual extraction of features beforehand which may change from dataset to dataset or depending on the disease application. Deep learning, on the other hand, have the ability to process the raw signals and classify data without requiring any domain knowledge or manually extracted features but lacks a good understanding and interpretability. This chapter will discuss different techniques of machine learning including features extraction and selection methods from filtered signals and classification of these selected features for clinical applications. We have also discussed two case studies i.e., epilepsy and schizophrenia detection. These case studies use an architecture which combines deep learning with traditional ML techniques and compare their results. Using this hybrid model, an accuracy of 94.9% is obtained based on EEG signals obtained from epileptic and normal subjects, while an accuracy of 98% accuracy is achieved in schizophrenia detection using only three EEG channels. The latter result is significant as it is comparable to other state of art techniques while requiring less data and computational power.