CLASSIFICATION OF MOTOR IMAGERY EEG SIGNALS USING ENSEMBLE AND DEEP LEARNING MODELS

Abstract

Electroencephalography (EEG) has become a prominent ongoing research topic in deep learning due to its complex nature and great relation to human brain activities. A lot of research is ongoing based on both machine learning and deep learning methods to figure out the usage of EEG signals to help both medical and non-medical areas. Understanding EEG can be used for a variety of tasks such as disorders diagnosis such as epilepsy, movement disorder, depression, schizophrenia, emotion detection, motor functions brain patterns, and many other non-medical usages such as virtual reality and gaming, identification and authentication, vehicle control, etc. EEG data contains voltage with respect to time of value in micro-volts in the brain, which is measured using devices such as BCI2000. EEG has become popular due to its success with deep learning models. It has been proven to give good results on various state-of-the-art algorithms. There are still a lot of areas to be researched and explored which can help us to solve problems of both medical and non-medical nature. Through the study of neural signals, brain-computer interfaces (BCI) allow for the study of direct communication between humans and other external devices. The BCI includes consecutive tasks like signal acquiring, feature extraction, training, and extracting results. The focus of our work is to use the ensemble method to improve overall accuracy. Unweighted averaging is used as a method to ensemble the deep learning models. The classification is done for four classes of motor imagery function-left hand, right hand, both hands, and both feet taking the optimal number of electrodes based on Granger causality and electrode pairing.