Automatic Detection of Seizures Using Extreme Learning Machine with a Single Feature

Automatic seizure detection is of great importance in clinical practice of epilepsy. This paper presents a classification system based on discrete wavelet transform (DWT) and the extreme learning machine (ELM) for epileptic seizure detection by distinguishing ictal and interictal electroencephalogram (EEG) signals. The original EEG signal is first decomposed by Daubechies order 4 wavelet into several sub-bands. Then, standard deviation, log of amplitude, and quartiles are calculated for the original and decomposed signals to construct feature vectors. Different combination of these features are fed into ELM and support vector machine (SVM). After comparing different combination strategies, we find that, using ELM, even with a single feature (standard deviation) from a single sub-band signal (4-8Hz), one can obtain a satisfactory classification result, which remarkably reduce the computational complexity and make the detection system more practical.