Minimum Feature Selection for Epileptic Seizure Classification using Wavelet-based Feature Extraction and a Fuzzy Neural Network

Abstract

This paper proposes a method that uses a wavelet transform (WT) and a fuzzy neural network to select the minimum number of features for classifying normal signals and epileptic seizure signals from the electroencephalogram (EEG) signals of people with epileptic symptoms and those of healthy people. WT was used to select the minimum number of features by creating detail coefficients and approximation coefficients from EEG signals. 40 initial features were obtained from the created wavelet coefficients using statistical methods, including frequency distributions and the amounts of variability in frequency distributions. We obtained 32 minimum features with the highest accuracy from the 40 initial features by using a non-overlap area distribution measurement method based on a neural network with weighted fuzzy membership functions (NEWFM). NEWFM obtains the bounded sum of weighted fuzzy membership functions (BSWFM) for the 32 minimum features to identify fuzzy membership functions for the 32 features. Using these 32 minimum features as inputs in the NEWFM resulted in a performance sensitivity, specificity, and accuracy of 99.67%, 100%, and 99.83%, respectively.