An Approach for Prediction of Epileptic Seizures from EEG Signals Through Analysis of Slow Components

An accurate and compact epileptic seizure prediction system would be significant to intractable patients. This paper researches on the forecasting of epileptic seizures based on scalp electroencephalography (EEG) recordings. The EEG signals can be regarded as a time series generated by a dynamic process. Considering the dynamic variation of the process, this paper puts forwards a novel approach employing slowness analysis that derives some latent variables. A Discriminative Slow Component Analysis algorithm is introduced to derive predictive features from EEG signals. Moreover, discrete wavelet transform is used to form epileptiform sub-waves that bring more predictive information. Finally, a cost-sensitive linear support vector machine is adopted to classify the preictal and interictal signals. Evaluated on a public EEG dataset, i.e., the CHB-MIT scalp EEG database, the proposed approach exhibited acceptably satisfactory performance.