Performance analysis of machine learning algorithms for classification of cardiac arrhythmia

Abstract

In each and every year, Cardio Vascular Diseases (CVD) produces newlineapproximately several million deaths around the world. Electrocardiogram newline(ECG) is meant as the electrical activity recording of heartand#8223;s cardiac muscle. newlineCardiac Arrhythmias (CA) are generally represented as the heart rate newlineabnormalities and the heart rhythm abnormalities. The CA signals are newlineobtained from Massachusetts Institute of Technology-Beth Israel Hospital newline(MIT-BIH) Arrhythmia physionet database. The raw ECG signal which is newlineacquired from MIT-BIH arrhythmia database will be influenced with huge newlinenoise signal components. To remove the noise signals and the baseline newlinewander from the ECG wave, preprocessing mechanism is being done. After newlinethis stage, the dimensionality reduction and the feature extraction is being newlinecarried out from the acquired denoised signal. Predominant features will lend newlineexact and useful information regarding the cardiac arrhythmias. In this newlineresearch, preprocessing is carried out by Discrete Wavelet Transform newlinetechnique and dimensionality reduction and feature extraction is carried out newlineby Principal Component Analysis and Independent Component Analysis built newlinewith non-parametric power spectral estimation respectively. After the newlineextraction of features using ICA, it is processed as an input to the classifier to newlinedetermine the classification of cardiac arrhythmia. In this research four newlinevarious types of classifiers are being utilized. They are Genetic Algorithm- newlineSupport Vector Machine (GA-SVM), Particle Swarm Optimization-Support newlineVector Machine (PSO-SVM), Artificial Feed Forward Neural Network using newlineBack Propagation newline