Chapter 1 - Prototype algorithm for three-class motor imagery data classification: a step toward development of human–computer interaction-based neuro-aid

Abstract

The quantification of spectral components of electroencephalogram (EEG) signals plays a crucial role in various clinical and scientific real-time applications for brain–computer interface (BCI) systems. For effective implementation of such systems identification of motor imagery (MI) signals from the frequency patterns obtained from various brain oscillations play a key role. The main objective of this chapter is to perform a comparative analysis of various feature extraction techniques and classification algorithms that are best suited for obtaining the MI signals. Linear features such as power spectral density, band power, as well as mean and maximum power of MI signals are analyzed along with some nonlinear features such as correlation coefficient and approximate entropy. An attempt has been made to check the different linear and nonlinear features from the EEG signals. These features are then fed to different classifiers such as support vector machine (SVM), linear discriminant analysis (LDA), and k-nearest neighbors (k-NN) to determine the best features providing maximum separability between different classes. In order to further enhance the classification accuracy, particle swarm optimization (PSO)-based SVM classifier is implemented. Band power features exhibit best performance over all other feature extraction techniques having a classification accuracy of 87.78% and 94.44% using SVM and PSO-SVM classifiers, respectively, for the BCI dataset. The case studies and results validate the efficacy of the proposed approach for real-time neuro-aid applications such as in moving a wheelchair or a robotic arm and other prosthetic devices. Also as the existing algorithms sometimes prove to be computationally slow, therefore, optimization techniques used may aid in enhancing the speed and accuracy of the systems.