An optimal structure of multilayer perceptron using particle swarm optimization for the prediction of cardiac arrhythmias

Abstract

In this paper, a novel approach for prediction of the cardiac arrhythmia classes is suggested using the Particle Swarm Optimization (PSO) and Multi-Layer Perceptron (MLP). The structure of MLP is optimized to improve the performance of classifier for the prediction of heart diseases. The linear and nonlinear methods are used to extract the features from the heart rate time series. Non-linear and linear parameters such as Largest Lyapunov Exponent (LLE), Spectral Entropy (SE), Hurst exponent (H), SD1/SD2 ratio, normalized Low Frequency (nLF) and High Frequency (nHF) power components are obtained from eight types of heart arrhythmias. Cardiac rhythms including Normal Sinus Rhythm (NSR), Premature Ventricular Contraction (PVC), Right Bundle Branch Block (RBBB), Left Bundle Branch Block (LBBB), Atrial Fibrillation (AF), Sinus bradycardia (SB), second degree Block (B) and Atrial flutter (A) are acquired from MIT-BIH arrhythmia database. The testing and training datasets for MLP Classifier are prepared from RR interval time series. Experimental results demonstrate that the proposed method based on PSO and MLP offers a significant improvement over the MLP which has manually parameter setting.