Classification of seizure and seizure-free EEG signals using multi-level local patterns

This paper introduces a new discriminant feature-Multi-level local patterns (MLP) for classification of seizure and seizure-free electroencephalogram (EEG) signals. The proposed approach employs Empirical mode decomposition (EMD) in order to decompose non-stationary EEG signals into intrinsic mode functions (IMFs). Multi-level local patterns are computed for each of these IMFs by performing comparisons in the local neighborhood of a sample value of the signal. Finally, a feature set is formed by computation of histograms of MLPs. In order to classify the EEG signal based on these features, we employ the nearest neighbor (NN) classifier, which utilizes scores computed from matching of histogram features of MLPs to determine the category of the EEG signal. Experimental evaluation of this approach on publicly available EEG dataset yielded improved classification accuracies as compared to the existing approaches in the literature. The best average classification accuracy of the proposed approach is 98.67%, which demonstrates the discriminatory capability of the proposed multi-level local patterns.