SVM classification study of code-modulated visual evoked potentials

We present a study of a support vector machine (SVM) application to brain-computer interface (BCI) paradigm. Four SVM kernel functions are evaluated in order to maximize classification accuracy of a four classes-based BCI paradigm utilizing a code-modulated visual evoked potential (cVEP) response within the captured EEG signals. Our previously published reports applied only the linear SVM, which already outperformed a more classical technique of a canonical correlation analysis (CCA). In the current study we additionally test and compare classification accuracies of polynomial, radial basis and sigmoid kernels, together with the classical linear (non-kernel-based) SVMs in application to the cVEP BCI.