A New Classification Method Based on KF-SVM in Brain Computer Interfaces

Abstract

This paper proposes a novel classification method named KF-SVM (Kernel Fisher, Support Vector Machine), which is used for the EEG (Electroencephalography) classification of two classes of imagery data in BCIs (brain-computer interfaces). This method combines the kernel fisher and SVM. Its detailed process is as follows: First, the CSP (Common Spatial Patterns) is used to obtain features, and then the within-class scatter is calculated based on these features. The scatter is added into the RBF (Radical Basis Function) kernel function to construct a new kernel function. The obtained new kernel is integrated into the support vector machine to get a new classification model. The KF-SVM may overcome the following defects of the SVM: 1) the SVM maximizes the classification margin without considering within-class scatter. 2) The classification surface of the SVM between two types of EEG data only depends on boundary samples and misclassified samples. To evaluate effectiveness of the proposed KF-SVM method, the data from the 2008 international BCI competition and experiments of our laboratory are processed. The experimental result shows that the proposed KF-SVM classification algorithm can well classify EEG data and improve the correct rate of EEG recognition in BCIs.