Classification of ECG Arrhythmia by Optimizing Multi-layer Perceptron Neural Network Using Genetic Algorithm

Abstract

An Electrocardiogram (ECG) graphically records changes in electrical potentials between different sites on the skin due to cardiac activity. ECGs are inexpensive and non-invasive means to observe the heart’s physiology. The heart’s electrical activity is a depolarization and depolarization sequence. ECGs help in identifying cardiac arrhythmia because they have diagnostic information. ECG arrhythmia detection accuracy improves by using machine learning and data mining methods. Genetic Algorithm is an optimization technique trying to replicate natural evolution where individuals with best characteristics adapting to the environment are likely to reproduce and survive. This study proposes multi-layer perceptron neural network optimization using Genetic Algorithm (GA) to classify ECG arrhythmia. Symlet extracts RR intervals from ECG data as features while symmetric uncertainty assures feature reduction. GA optimizes learning rate and momentum. Experimental results show that the proposed optimized neural network achieved classification accuracy and average precision of 96.93% and 96.92% average recall.