Stationary Wavelet Transform for Automatic Epileptic Seizure Detection

Gebremichael Shiferaw, Adane Letta Mamuye, Marco Piangerelli

Abstract: Visual detection of epileptic seizure from EEG signal is being inefficient and time consuming. Computational EEG signal analysis techniques were then used in the diagnosis and management of epileptic seizures. In this study, we compared the performance of Discrete Wavelet Transform (DWT) and the Stationary Wavelet Transform (SWT) decomposition techniques with 22 wavelet functions (Coiflets (coif), Daubechies (DB) and Symlets (Sym) families) using support vector machine classifier. We used multichannel EEG dataset of the University of Bon Epilepsy Center. From this dataset, five statistical wavelet features: max, min, average, mean of absolute and standard deviation were extracted. In all of the wavelet functions except three, in the Coiflets family, the experimental result showed that SWT achieved better classification accuracy than DWT. SWT and DWT decomposition techniques registered 99.5% and 97.5% highest classification accuracies, respectively.