Classification of Motor Imagery Tasks using Phase Space Reconstruction and Empirical Mode Decomposition

Motor imagery (MI) based brain computer interface (BCI) is an assistive device for the motor disabled people. But it has limited applications due to its lower classification performance. To enhance the performance, this paper introduces an efficient method for the detection of both left and right hand MI tasks of the subject using phase space reconstruction (PSR) and empirical mode decomposition (EMD). First, EMD was employed on MI-EEG signals to obtain a set of band limited functions called as intrinsic mode functions (IMFs). To study the MI activities, the IMFs whose main frequency lies between 8- 30 Hz (i.e. sensorimotor frequency band) were selected. On the other hand, PSR was applied to the selected IMFs followed by the extraction of MI features. At last, the significant features (p¡0.05) extracted from one-way analysis of variance (ANOVA) were fed into different machine learning models such as logistic regression (LR), Naive Bayes (NB) and support vector machine (SVM) to detect MI tasks. The proposed method and the classifiers were tested on BCI competition 2003 MI dataset. The results show that the SVM improved the classification accuracy upto 4.27% with better performance (i.e. % CA=96.67%, K=0.93 and Auc=0.96) and outperformed the existing methods reported in the literature (maximum % CA=92.40%).