Comparing performance of multi-class classifiers for grasping patterns from EEG data

Abstract

In this paper, we construct 6 types of multi-class classifiers and compare the performance of them. In the feature extraction phase, CSP filter (common spatial pattern filter) and PCA (principle component analysis) and LDA (linear discriminant analysis) are used. In the classification phase, kNN (k-nearest neighbor) method and SVM (support vector machine) are used. As a result, the best classifier for discriminating grasping patterns was CSP/kNN classifier (We set the selective number k as 1). The best classification rate for 9-class power-type grasping patterns of CSP/kNN classifier was 48% and the classification rate for 7-class precision-type grasping patterns of CSP/kNN classifier was 40%.