Cardiac Arrhythmia Detection by ECG Feature Extraction: A Machine Learning Based Approach

Abstract

ECG beats are vital for reducing fatalities from CVDs by enabling arrhythmia detection through intelligent systems, which provide crucial cardiac insights to specialists. However, challenges such as noise, heartbeat instability, and imbalance affect the accuracy and speed of these systems. Accurate diagnosis in quick time is essential for proper treatment and patient recovery. This study focuses on enhancing the precise diagnosis of various CVD types by analyzing arrhythmias in ECG signals of the heartbeats. We developed a deep learning based arrhythmia detection system that utilizes discrete wavelet transformation during pre-processing of the signals and the SMOTE oversampling algorithm to deal with the class imbalance problem. Our classifier integrates a Convolutional Neural Network (CNN) for spatial pattern detection with a Bidirectional Long Short-Term Memory (BLSTM) network for temporal dependency identification. We trained and evaluated our system using the MIT-BIH Arrhythmia Dataset.The evaluation results demonstrate that our method, after 50 training epochs, achieves high accuracy in different categories: 99.65% for class F, 99.37% for class V, 98.45% for class N, 99.10% for class S, and 99.82% for class Q. This proposed deep learning based system can be employed for the automatic diagnosis of arrhythmia and assist the CVD specialists in accurate diagnosis.