Optimal Feature Combination Using SVM Algorithms for Brain-computer Interface

This paper aims to establish a four-class classification brain-computer interface model with the optimal combination of features. By exploring the influence of feature vector combinations in different time windows on brain-computer interface (BCI) classification, the optimal 2-feature combination could be get. Four types of movement paradigms were studied based on the concentration of oxygenated hemoglobin (HbO) in the brain motor region. A total of 10 volunteers completed a set of movement patterns which included hand griping and arm lifting of two upper limbs. After extracting signal mean(SM) and signal slope(SS) eigenvectors, the “one-versus-one” SVM algorithm was used for classification. The highest average classification accuracy rate reached 80.48%, and most of the participants got the highest classification accuracy in the combination time window of SM feature (0~5s) and SS feature (5~7s). In the case of the same time window of SM feature, the classification accuracy increased gradually with the selection of 3~5s, 4~6s, 5~7s respectively in the time window of SS feature.