EEG classification based on variance

Brain computer interface (BCI) establishes a communication between human brain and an external assistive device via a computer. This communication system is mainly used for physically challenged individuals as well as it can be used to command robot to do task depending on mental thought. In this paper, we are working on the Electroencephalogram (EEG) signals based BCI system because the EEG signals having high temporal resolution and it is non-invasive. The objective of the paper is to enhance the classification accuracy for the movement imagery under the unsupervised learning. We have selected four preprocessing conventional filters namely Chebyshev filter, Butter worth filter, FIR bandpass filter and Elliptic filter to compare the performance in terms of classification accuracy. The filtered data is segmented and statistical parameter variance has been calculated in time domain. The difference of variance of the channels C3 and C4 has been taken as a feature. The Fisher linear discriminant analysis (FLDA) has been used to classify the feature matrix. The Elliptic filter achieves 84.3% classification accuracy for training data and 86.45% for testing data between the left and right hand movement imagination.