The study demonstrate that proposed IHMv classifier can distinguish more output classes than other researches with less number of electrodes while maintaining similar accuracy classification. Therefore, our system can be applied to BCI system to create controlling commands to peripheral devices or computer application through motor brain waves. | Journal of Science & Technology 123 (2017) 048-053 Classification of Three Class Hand Imagery Movement with the Application of 2-Stage SVM Model Pham Phuc Ngoc, Pham Van Binh* Hanoi University of Science and Technology, No. 1, Dai Co Viet, Hai Ba Trung, Hanoi, Viet Nam Received: May 26, 2016; Accepted: November 03, 2017 Abstract Classification of multiple motor states of human brain waves will improve the applicable ability to brain – computer interfacing system and neuroprosthesis. In this research, we aim at extending classifier ability to three class imagery movement states including: left hand imagery movement, right hand imagery movement and rest state. We propose to implement new classifier using combination of discriminated features for input feature vector and classifying model based on 2-stage SVM. For the feature vector construction, ANOVAbased feature selection method is proposed to use which resulting in 14% reduction of number of features required for the process. The avarage classification accuracy of our proposed system is achived with 80,75% when evaluated on Physionet dataset. The study demonstrate that proposed IHMv classifier can distinguish more output classes than other researches with less number of electrodes while maintaining similar accuracy classification. Therefore, our system can be applied to BCI system to create controlling commands to peripheral devices or computer application through motor brain waves. Keywords: EEG, 3 - class imagery hand movement, 2-stage SVM model, Wavelet, motor cortex, ANOVA based time-frequency feature extraction method. Those features are potential used to create an input feature vector for three – class classification model. Those features are generated from only 2 channel C3, C4 which located over human brain motor area have shown their ability to pick up hand motor information [10],[8]. To remove redundant features for our classification problems, in this study, we also propose ANOVA – based feature .
