Comparative study of two movement identification strategies on Brain-Computer Interface motor task

Support Vector Machine (SVM) is a state-of-art machine learning algorithm broadly used classification and learning tasks. It has been recently used on Brain-Computer Interface (BCI) systems to discrimination between motor tasks of electroencephalography (EEG) signals. However, there are different possible strategies to implement the SVM classifier. In this paper1 it is compared two SVM strategies to classify a motor movement task: 3-class and binary, hierarchical SVM. One healthy subject was submitted to an offline BCI experiment and the results show that the binary classifier reports in average higher classification accuracies than the 3-class SVM (68.47% against 53.07%). The difference between both strategies is statistically significant (p-value < 0.001; for 0.05 significance level).