Surface EEG based epileptic seizure detection using wavelet based features and dynamic mode decomposition power along with KNN classifier

The seizure is defined as the sudden synchronous activity of the number of neurons resulting in abnormal body symptoms. This paper proposes a technique for the auto-detection of epileptic seizures using an online surface EEG database. The features were extracted for every 5 seconds window from the online surface EEG signal. Authors used dynamic mode decomposition power feature calculated from the multichannel EEG signal and Power spectral density, variance, and Katz fractal dimension features evaluated from wavelet packet decomposition coefficients for seizure detection. The K-nearest neighbor (KNN) classifier was used for classification. The KNN classifier was trained separately for each signal feature. The proposed system achieved good classification accuracy in seizure detection using a simple KNN classifier. The approach is further verified using the All India Institute of Medical Sciences (AIIMS) Patna seizure database and online seizure EEG database collected from neurology and sleep center, Hauz Khas, New Delhi. Different types of seizures were considered for validation of the model. KNN classifier-based approach achieved 98.99%, 99.69%, and 96.25% classification accuracy in detecting seizures from the online surface EEG seizure database, AIIMS Patna EEG seizure database, and online seizure database collected from neurology and sleep center, Hauz Khas, New Delhi. Support vector machine classifier was further evaluated for accuracy in seizure detection from the EEG signal collected at neurology and sleep center, Hauz Khas, New Delhi available online and achieved 95.5% accuracy in preseizure-seizure EEG segment classification and 96.5% accuracy in interseizure-seizure EEG segment classification. SVM Radial Basis Function (RBF) kernel based approach achieved highest accuracy compared to linear and polynomial kernel based approach.