Evolution of time-domain feature for classification of two-class motor imagery data

Abstract

Brain–computer interface system that provides communication between the brain and computer by transforming the raw signals into the command or communication signal. Its emerging technology in the rehabilitation engineering for people who are physically challenged, it improves their everyday activities independently. Motor imagery is the fundamental part, which consists of imagination of activity such as hand and feet movement. Classification of motor imagery task is a challenging task; it not only depends on the classification algorithms but also the quality of signals (input) is also important. For this purpose, feature extraction methods are used, which highlight the property of the signal that differentiates this particular signals from others. Performance of motor imagery-based BCI directly relies on the efficacy of feature extraction and classification algorithms. In this paper, time-domain features such as mean absolute value, zero crossing, slop sign change, and waveform length are considered for classification of two-class motor imagery EEG data. Combination of two mental tasks, right hand, feet movement has been considered for classification with ten different classifiers for three subjects. It is found that the author got accuracy lies between 55 and 59% which is relatively less for designing the efficient BCI. So in future work, performance improvement will be done through applying some filter, artifact removal, and thresholding method.