Optimization-enabled deep convolutional neural network with multiple features for cardiac arrhythmia classification using ECG signals

Abstract

Arrhythmia is a heart disorder because of irregular electrical activity of the heart. Generally, an electrocardiogram (ECG) is a device used by a medical specialist to determine heart abnormalities or arrhythmias. The conventional classification approaches utilized a significant amount of time and attained only minimum accuracy. Thus, this work develops an effectual approach; named Chameleon-Sparrow Search Algorithm (CsSA) to train the Deep Convolutional Neural Network (Deep CNN) using ECG signals to classify arrhythmia. By utilizing the Daubechies wavelet filter, the pre-processing is done. Subsequently, wave components are detected by utilizing a multi-resolution wavelet-based model for eight-level decomposition employed for ECG signal. Then, features from DWT, auto-regressive, Empirical mode decomposition (EMD) features, and Variational mode decomposition (VMD) features are extracted. Moreover, Chameleon Swarm Algorithm is employed to filter representative features for arrhythmia diagnosis. Finally, the arrhythmia classification is progressed by proposed Chameleon-Sparrow Search approach-based Deep CNN (CsSA-based Deep CNN), where Deep CNN is trained by exploiting proposed CsSA. CsSA is proposed by collaboration of Sparrow Search Algorithm (SSA) and Chameleon Swarm Algorithm (CSA). The performance is estimated and compared with that of existing strategies regarding specificity, accuracy, and sensitivity. From analysis, it can be exhibited that CsSA-based Deep CNN obtains the highest value of accuracy, sensitivity, as well as specificity at a rate of 0.947, 0.929, and 0.964 respectively.