Prediction of Cardiac Arrhythmia Using Multi Class Classifiers by Incorporating Wavelet Transform Based Features

Abstract

Timely diagnosis and earlier detection of the dangerous heart conditions will reduce the mortality rate and save life of the patient. For that, it is necessary to automate the classi?cation and prediction of Cardiac Arrhythmia. Raw ECG signal is extracted from the MIT-BIH Arrhythmia database, followed by preprocessing and feature extraction using wavelet transform method. Further the extracted features are used for the classification of four different cardiac arrhythmias such as Bradycardia, Tachycardia, Left and Right Bundle Branch Block. Comparative study on the five different classifiers namely Decision trees, Support Vector Machine (SVM), Discriminant Analysis, k-Nearest Neighbor Classifiers (KNN), Ensemble Classifiers, and its variants are experimented in the proposed work. Among these, the weighted KNN classifier gives higher accuracy (90.3%) and prediction speed (10,000 observations per second) with reduced training time (4.329 seconds), compared with the existing state of the art methods. The prediction speed is 10,000 numbers of observations per second which identifies the heart problem earlier, and so appropriate treatment can be given to the patient. To further improve the classification accuracy, three optimizable classifiers namely Optimizable KNN, optimizable SVM, optimizable ensemble are used for the hyper parameter tunning and weight optimization. The optimizable SVM provides better perform (accuracy 93.4 %) among the three optimizable classifiers as well as the existing state of the art works. Therefore, the proposed work used for earlier Cardiac arrhythmia disease diagnosis and prognosis. Keywords: ECG, Cardiac Arrhythmia, Wavelet Transform, Multi class Classifiers, Decision trees, Support Vector Machine (SVM), Discriminant Analysis, k-Nearest Neighbor Classifiers, Ensemble Classifiers, Optimizable classifier.