Statistical Signal Processing and Machine Learning Based Diagnosis of Arrhythmia

The present work deals with the detection and classification of arrhythmia based on the analysis of an Electrocardiogram (ECG) signal. ECG signals have been collected from normal healthy persons and patients suffering from arrhythmia. Continuous Wavelet transform (CWT) and Discrete Wavelet transform (DWT) based statistical parameters are computed on the denoising ECG signals. In CWT analysis distinctive features have been computed for arrhythmia patients over normal healthy people. In this work Sinus Arrhythmia, Ventricular tachycardia (3 beats and 7 beats), Sinus bradycardia, Bidirectional ventricular tachycardia, Premature Ventricular Contractions types of Arrhythmia has been used. Clear differences in mean values and standard deviations for approximation and detail coefficients in DWT analysis have been noticed in both the cases. All these feature patterns and statistical analyses are used to diagnose arrhythmia. Then arrhythmia has been classified by different machine learning (ML) based techniques where maximum accuracy has been achieved in Random Forest and k-nearest neighbor (KNN) methods. Several cases have been considered to validate the proposed method where each and every cases result has obtained very much optimistic.