Classification Algorithms Employed by EEG-Based BCI - A Comparative Survey
Brain-Computer Interface (BCI) is a technology in cognitive science that maps a user’s neural signals to
commands that are further relayed to an output device in order to carry out the desired action. A variety of signals can be
acquired and analysed for BCI applications, however, we will be focusing on the Electroencephalographic (EEG) signals in
this survey. Fundamentally, a BCI system consists of signal acquisition, data preprocessing, extracting relevant features and
their classification. For the final classification module, a number of machine learning approaches such as Support Vector
Machines, Linear Discriminant Analysis, Naive Bayes, Decision Trees, k-NN and Random Forest have been used
traditionally. However, the focus is now shifting towards the more efficient deep learning techniques like Convolutional
Neural Networks, Deep Belief Networks and a combination of models, for classification. The neural network classifiers are
by and large seen to be favored over the one-size-fits-all strategies of the traditional machine learning classifiers which are
suitable for a wide range of solutions. In this survey, we present the major classification techniques employed over the years
in the research of EEG-based BCI and provide a comparative analysis of the same. We take the percentage accuracy as a
performance measure for comparison.
Keywords - Machine Learning, Deep Learning, Classification, Brain-Computer Interface, Electroencephalographic signals.