Combined LVQ neural network and multivariate statistical method employing wavelet coefficient for EEG signal classification

Abstract

Since that electroencephalogram (EEG) signals contains vital information about brain health, for better diagnosis analyzing EEG signals is important. This paper developed new classifier architecture using combined neural network (NN)-wavelet transformer (WT) and statistical methods to classification EEG signals. For increasing the accuracy and speed of classification, the exact classes are determined using WARD multivariate statistical methods and dendogram graph. Then discrete WT (DWT) and wavelet packet (WP) coefficients of EEG signals are applied to training of NN separately. For determining the effect of NN training method in results, two different supervised and unsupervised NN is selected: multilayer perceptron (MLP) and learning vector quantization (LVQ). Classification accuracy of LVQ-WT, LVQ-WP and MLP-WT methods is 95.67%, 97% and 98.67% respectively that show good ability of MLP-WT in classification.