Cardiac Arrhythmia multiclass classification using optimized FLS-based 3D-CNN

Abstract

Arrhythmia is the medical term for any irregularities in the normal functioning of the heart. Due to their ease of use and non-invasive nature, electrocardiograms (ECGs) are frequently used to identify heart problems. Analyzing a huge number of ECG data manually by medical professionals uses excessive medical resources. Consequently, identifying ECG characteristics based on machine learning has become increasingly popular. However, these conventional methods have some limitations, including the need for manual feature recognition, complex models, and lengthy training periods. This research offers a unique hybrid POA-F3DCNN method for arrhythmia classification that combines the Pelican Optimisation algorithm with fuzzy-based 3D-CNN (F3DCNN) to alleviate the shortcomings of the existing methods. The POA is applied to hyper-tune the parameters of 3DCNN and determine the ideal parameters of the Gaussian Membership Functions used for FLSs. The experimental results were obtained by testing the performance of five and thirteen categories of arrhythmia classification, respectively, on UCI-arrhythmia and the MIT-BIH Arrhythmia datasets. Standard measures such as F1-score, Precision, Accuracy, Specificity, and Recall enabled the classification results to be expressed appropriately. The outcomes of the novel framework achieved testing average accuracies after ten-fold cross-validation are 98.96 % on the MIT-BIH dataset and 99.4% on the UCI arrhythmia datasets compared to state-of-the-art approaches.