A Comparative Analysis of Advanced Deep Learning Techniques for Accurate Cardiac Arrhythmia Classification

The precise identification of cardiac arrhythmias facilitates accurate diagnosis and proper treatment, but the characterization process remains complex due to disturbances in ECG data signals along with skewed class frequencies and individual patient-specific variations. This study developed a deep learning framework, known as Penalty Regression Function-enhanced Deep Convolutional Neural Network (PRF-DCNN), as a comprehensive solution to cope with signal noise along with class imbalance and variations in patient data. The system starts by applying Correlation Factor-Based Extended Kalman Filtering (CF-EKF) for ECG signal denoising before allowing Ensemble Empirical Mode Decomposition (EEMD) to extract nonstationary features. The feature selection process along with the reduction of redundant characteristics uses the Frechet Fitness Rank Distribution-Anas Platyrhynchos Optimization (FFRD-APO method. The dataset is balanced by a Balanced Zero Noise GAN (BZNGAN) before Age-Weighted Average-Based Farthest First Clustering (AWA-FFC) refines the clustering process. The St. Petersburg INCART 12-lead ECG dataset was used to test the model, which obtained 99.53% accuracy, 99.10% sensitivity, and 99.67% specificity. The proposed system outperforms current models, showing its capacity for dependable time-critical arrhythmia detection in medical environments.