A novel approach for extracting feature from EEG signal for mental task classification

In the last few years, plenty of works have been done by the research community of Brain-Computer Interface (BCI), which assists physically challenged people to communicate with the help of brains electroencephalograph (EEG) signals. An appropriate representation of these brain signals, for mental task classification, in terms of relevant features is important to achieve high classification performance. There are vast numbers of choices of transformation techniques from one domain to another domain for non-linear and non-stationary brain signals. In spite of the availability of these vast transformations, it is difficult to identify the most appropriate and relevant features for classification of mental task. In this work, we propose a new approach for extracting features from EEG signals by estimating parameters based on statistical, complexity, spectral and entropy of the signal for mental task classification. To validate our approach, experiments has been performed on the publically available dataset of Keirn & Aunon [1]. Also, to capture the sensitivity of classification, performance to different classifiers, four well-known classifiers has been used in our experiments and Friedman statistical test has been performed to rank various methods. Experimental results show that the performance of classification increases using the proposed approach of feature extraction.