Epileptic seizure detection using GARCH model on EEG signals

In this paper, we suggest a new classification method for Electroencephalogram (EEG) signals, based on statistical modeling of wavelet coefficients. First, we demonstrate that Generalized Autoregressive Conditional Heteroscedastcity (GARCH) effect exists in wavelet coefficients of EEG signals utilizing Kolmogorov-Smirnov (K-S) test. By using GARCH model on wavelet coefficients, correct classification rate (CCR) is improved. First, the EEG signals are decomposed into frequency sub-bands, using discrete wavelet transform (DWT). Then a set of statistical features are extracted from each subband to represent the distribution of wavelet coefficients. Also we calculate GARCH variance series from wavelet coefficients. We use linear discriminant analysis (LDA) selection techniques for feature selection and reduction. These result feature vectors are fed to multilayer perceptron (MLP) classifier for EEG classification.