A Review on Adaptive Classifiers for BCI Classification

Abstract

A Brain-Computer Interface (BCI) aims at providing a way for controlling external devices through the utilization of brain signals. One of the challenges in electroencephalography (EEG)-based BCI is to adjust the brain signal decoder to detect a user's intention as accurately and efficiently as possible, as EEG signals are non-stationary. Therefore, adaptive classification, an approach to adapt to the changes of the EEG signals, would be effective in overcoming this problem. This paper provides a review of the representative adaptive classifiers used in BCI, and it can be divided into four categories: adaptive linear discriminant analysis, adaptive support vector machine, adaptive Bayesian classifiers and adaptive Riemannian Geometry-based classifiers. Besides, the pros and cons of these adaptive classification algorithms are further described.