Principal component analysis-based neural network with fuzzy membership function for epileptic seizure detection

Abstract

A hybrid principal component analysis (PCA)-based neural network with fuzzy membership function (NEWFM) is proposed for epileptic seizure detection. By combining PCA and NEWFM, the proposed method improves the accuracy in epileptic seizure detection. The PCA is used for wavelet feature enhancement needed to eliminate the sensitivity of noise, electrode artifacts, or redundancy. NEWFM, a model of neural networks, is integrated to improve prediction results by updating weights of fuzzy membership functions. A dataset made up of 5 sets, each consisting 100 single EEGs segments, is employed to evaluate the proposed system's performance. Based on the experiments, the prediction results show an accuracy rate of 98.29% for epileptic seizure classification while in the best cases the accuracy reaches 99.5% for the `normal' (Z-S) seizure classification task.