Epileptic State Classification for Seizure Prediction with Wavelet Packet Features and Random Forest

The epileptic state classification for preictal prediction of seizure is essential for preventing the damages caused by seizure. Existing research mainly focus on the epileptic seizure detection while few of them investigate the preictal state prediction. In this paper. a novel electroencephalograph signal (EEG) based prediction algorithm for the preictal state of seizure is developed. The one hour preictal state before the seizure is divided into three non-overlapped segments and a multi-class classification is then performed for preictal state prediction. The wavelet packet based features (WPFs), including the subband energy ratio and three wavelet entropies, are extracted for the multi-channel EEG signal representation. The Random Forest algorithm (RF) is adopted for the preictal state prediction. Experiments on the benchmark CHB-MIT EEG database are conducted for performance analysis. Comparisons with methods based on popular EEG features and classifiers are provided to show the superiority of the proposed WPFs+RF based preictal state prediction algorithm.