An Improved Support Vector Machine Classifier for EEG-Based Motor Imagery Classification

  • Authors:

  • Hui Zhou;Qi Xu;Yongji Wang;Jian Huang;Jun Wu

  • Affiliations:

  • Key Laboratory of Image Processing and Intelligent Control, Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan, China 430074;Key Laboratory of Image Processing and Intelligent Control, Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan, China 430074;Key Laboratory of Image Processing and Intelligent Control, Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan, China 430074;Key Laboratory of Image Processing and Intelligent Control, Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan, China 430074;Key Laboratory of Image Processing and Intelligent Control, Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan, China 430074

  • Venue:
  • ISNN 2009 Proceedings of the 6th International Symposium on Neural Networks: Advances in Neural Networks - Part II
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Electroencephalogram (EEG) recordings during motor imagery tasks are often used as input signals for brain-computer interfaces (BCIs). We analyze the EEG signals with Daubechies order 4 (db4) wavelets in 10 Hz and 21Hz at C3 channel, and in 10 Hz and 20 Hz at C4 channel, for these frequencies are prominent in discrimination of left and right motor imagery tasks according to EEG frequency spectral. We apply the improved support vector machines (SVMs) for classifying motor imagery tasks. First, a SVM is trained on all the training samples, then removes the support vectors which contribute less to the decision function from the training samples, finally the SVM is re-trained on the remaining samples. The classification error rate of the presented approach was as low as 9.29 % and the mutual information could be 0.7 above based on the Graz BCI 2003 data set.