Smart Arrhythmia Detection Using Single Lead ECG Signal and Hybridized Deep Neural Network Model

Abstract

Arrhythmia is an irregular electrical activity of the heart that needs to be treated quickly and promptly to avoid the risk of cardiac failure and stroke. Signal processing utilizing Electrocardiogram (ECG) signals continues to be the gold standard for detecting cardiac abnormalities. However, the low classification accuracy and lack of labeled ECG data might seriously impair the existing algorithm's overall performance. To address the drawbacks of the existing techniques, the proposed research utilizes a deep learning model formulated utilizing the cephalous wolf optimization-based deep neural network model (CWO opt NN) for effective arrhythmia detection. The proposed model leverages the characteristics of a single lead ECG database to retrieve the input data initially. Next, the signal is preprocessed by adopting the window sliding approach to eliminate any potential noise. In addition, the extracted time-domain features, frequency domain features, geometrical features, CSI-CVI features, wavelet features, and statistical features, aid in boosting the accuracy of arrhythmia detection. To accurately identify arrhythmia, the developed model explores the Neural Networks for learning the cardiac cycles effectively. Specifically, cephalous wolf optimization, developed by the typical hybridization of the cephalous wolf and wolf hawk, is essential to the research's relevance since it allows for the successful identification of arrhythmia by fine-tuning the classifier's weights and bias. Considering the achievement rates for arrhythmia identification at training percentage 80, the F1-score is 96.10%, precision is 97.08%, and recall is 95.14% respectively, similarly based on the k-fold 8, F1-score is 96.10%, precision is 96.80%, and recall is 94.86% respectively.