Comparison of artificial neural network and support vector machine classifications for fNIRS-based BCI

Abstract

In this paper we analyze and compare the performance of support vector machine (SVM) and artificial neural network (ANN) for classification of fNIRS signals. fNIRS signals due to mental arithmetic and mental counting are acquired from the prefrontal cortex of ten healthy subjects. After preprocessing and filtering, SVM and ANN classification is performed on the same feature set - mean and slope of the changes in concentration of oxy-hemoglobin. Although no significant difference in the average classification accuracies, obtained using SVM and ANN, is observed (p = 0.2); it is noted that the standard deviation of classification accuracies using ANN is significantly higher than that of SVM. Furthermore, the computational speed of SVM is significantly higher than that of ANN. It is concluded that SVM offers stable classification accuracies and fast computation as compared to ANN.