Feature analysis of epileptic EEG using nonlinear prediction method

We propose a feature extraction method based on the Volterra autoregressive model's prediction power and the data's predictability for the EEG signals to automatically detect the epileptic EEG signals from the EEG recordings. The method of determining the embedding dimension based on nonlinear prediction is applied to choose the embedding dimension of the EEG data. The proposed feature extraction method is used to extract the feature for three groups of EEG time series composing epileptic seizure. We divide the EEG data into segments, and respectively compute the feature values of each segment, where the length of data segment respectively takes the value of 250, 500, 1000 points. To investigate the robustness of our method under noises, we also analyze the three EEG time series with additive white Gaussian noise. The experiment results show that the feature values extracted with the proposed method could obviously distinguish the epileptic EEG signals from the normal EEG signals. The proposed method is effective for short time series, insensitive to the length of data segment, and robust to the additive white noise, and it could differentiate the epileptic EEG from the normal EEG when the signal-to-noise ratio is low.