Accuracy of Classification Algorithms Applied to EEG Records from Emotiv EPOC+ Using Their Spectral and Asymmetry Features

Abstract

To develop a good BCI, it is necessary to take into account what features can be extracted and what classification algorithm can be used. In this manuscript, a cross-validation method is used to compare different classification algorithms (SVM, KNN, discriminant analyses and decision trees) as applied to EEG records obtained by a non-invasive wireless electroencephalograph (Emotiv EPOC+). The features used in the classification algorithms are the power spectrum of the signal and the hemispheric asymmetry. The used experimental paradigms (e.g. motor imagery) are designed to be used with reduced mobility people, because the aim is to develop a BCI to control an external device such as a wheelchair or a prosthesis.