A Hybrid Particle Swarm Optimization and Neural Network with Fuzzy Membership Function Technique for Epileptic Seizure Classification

A hybrid particle swarm optimization (PSO) integrating neural network with fuzzy membership function (NEWFM) technique is proposed for epileptic seizure classification tasks based on brain electroencephalography (EEG) signals. By combining PSO and NEWFM, the proposed method obtains the optimal parameters from the EEG data training required to achieve the best accuracy in disease diagnosis. NEWFM, a model of neural networks, is expected to improve the accuracy by updating weights of fuzzy membership functions. The PSO, a swarm-inspired optimization algorithm, is used to obtain the optimal parameters from the NEWFM. A standard dataset comprising of 5 sets of epileptic seizure detection data, each consisting 100 single EEGs segments is employed to evaluate the proposed technique's performance. Based on the experiments, the classification results show that the best accuracy of Z-S classification task is 99.5% with the optimal parameters of alpha = 0.1 and beta = 0.1. For the ZNF-S classification task, the best accuracy is 97.73% with the optimal parameters of alpha = 0.1 or 0.2 and beta = 0.2. Similar results for the ZNFO-S classification task is 97.64% with the optimal parameters set at alpha = 0.1 or 0.2 and beta = 0.1.