Patient-Specific Automatic Seizure Detection Method from EEG Signals Based on Random Forest

Abstract

Epilepsy is an abnormal discharge in focal or whole part of brain, lasting a few seconds or minutes. The detection of epileptic seizure by the way of visual inspection is time-consuming, so the study for automatic seizure detection methods toward long-term electroencephalogram (EEG) recording is valuable. Due to the nonstationary characteristics of EEG signal, traditional analysis methods cannot achieve epilepsy diagnosis successfully. In this paper, we presented a method, namely, the patient-specific automatic seizure detection method, to identify epilepsy in EEG signals. First, a method based on time-domain and nonlinear characteristics is used to analyze the selected EEG segment and obtain the features of each segment. Then, these features are applied as the input of random forest to get classification result, concerning the existence of seizures or not. The accuracy of proposed method is 92.05%. Therefore, the proposed method is validated by using available dataset of online.