Performance enhancement of mental task classification using EEG signal: a study of multivariate feature selection methods

In the recent years, the research community has shown interest in the development of brain–computer interface applications which assist physically challenged people to communicate with their brain electroencephalogram (EEG) signal. Representation of these EEG signals for mental task classification in terms of relevant features is important to achieve higher performance in terms of accuracy and computation time. For feature extraction from the EEG, empirical mode decomposition and wavelet transform are more appropriate as they are suitable for the analysis of non-linear and non-stationary time series signals. However, the size of the feature vector obtained from them is huge and may hinder the performance of mental task classification. To obtain a minimal set of relevant and non-redundant features for classification, six popular multivariate filter methods have been investigated which are based on different criteria: distance measure, causal effect and mutual information. Experimental results demonstrate that the classification accuracy improves while the computation time reduces considerably with the use of each of the six multivariate feature selection methods. Among all the combinations of feature extraction and selection methods that are investigated, the combination of wavelet transform and linear regression performs the best. Ranking analysis and statistical tests are also performed to validate the empirical results.