Enhancing EEG-Based Brain–Computer Interface
Performance: A Review of Machine Learning Algorithms

Ahmed EL-Emam^1,2, Hossam El-Din Moustafa^3,4, W Mustafa⁵, Islam Ismael⁶, EL-Sayed M.El-Kenawy^,7,8,*

¹Higher Technological Institute for Applied Health Sciences, Department of Medical Equipment Maintenance, Dakahlia, Egypt

² Biomedical Engineering Department, Faculty of Engineering, Mansoura University, Egypt

³Electronics and Communications Engineering Department, Faculty of Engineering, Mansoura University, Egypt

⁴Dean of Faculty of Artificial Intelligence and information, Horus University, Egypt

⁵Department of Neurology, Mansoura Faculty of Medicine, Mansoura, Egypt

⁶Department of Electrical Engineering, Faculty of Engineering, Mansoura University, Egypt

⁷Delta Higher Institute of Engineering and Technology Department for Communications and Electronics Mansoura 35511, Egypt

⁸Applied Science Research Center. Applied Science Private University, Amman, Jordan

Emails: ahmedhamed@std.mans.edu.eg; hossammoustafa@mans.edu.eg; wesam010@mans.edu.eg; islam_m@mans.edu.eg; skenawy@ieee.org

Abstract

Brain-computer interface (BCI) systems based on electroencephalography (EEG) are applications that allow human-to-machine communication with intuitive (near-transparent) control, whose neural commands are decoded based on intentional movement. Recent research on the topic of machine learning (ML) has been able to greatly enhance the classification of the EEG-signals associated with the movement of the hands, head movements, and mobility movements of the eyes. The developments allow various utilization across assistive technologies, prosthetic control, and non-verbal communication. EEG, however, is highly non-stationery and noise-sensitive, so advanced preprocessing and optimization methods have to be applied to optimize performance in classification. This paper outlines an in-depth review of some of the most popular ML algorithms, i.e. support vector machines (SVMs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs), and optimization methods, i.e., genetic algorithms (GAs), particle swarm optimization (PSO), and transfer learning. We point out existing problems in the processing of EEG signals and suggest directions in the future that will improve the robustness, generalization, and real-time behavior of BCI.

Keywords: BCI; Machine Learning; Deep Learning; EEG; Optimization technique