ECG signal classification using MLP neural network with hybrid PSO-BP training algorithm

Electrocardiogram signals (ECG) are the important approach in heart activities monitoring and heart diseases diagnosis. In this paper evolvable multilayer perceptron neural network (MLPNN) is used for heartbeat pattern classification. Multilayer perceptron neural network (MLPNN) is formed of one or more hidden layers which can be trained by back propagation (BP) and/or evolutionary algorithms. MLPNN is trained by combination of particle swarm optimization (PSO) algorithm and back propagation (BP) algorithm, which is used to combine the PSO algorithm’s strong ability in global search and the BP algorithm’s strong ability in local search. MLPNN weights are optimized using particle swarm optimization algorithm. Heart signals are classified in five different classes by trained network according to association for the advancement of medical instrumentation. The inputs of neural network are features which have been extracted from ECG signals. The MIT-BIH arrhythmia database is used for simulation results. Classification accuracy of MLPNN for F signal 88.10%, N signal 96.49%, Q signal 73.68%, V signal 92.83% and S signal 95.93% is obtained. Simulation results show that proposed hybrid PSO-BP algorithm has better performance than BP algorithm in classification accuracy.

Keywords:

Back propagation , Electrocardiogram (ECG) , Heartbeat classifier , Particle Swarm Optimization