Enhanced ECG Classification Using Deep Convolutional Neural Networks

Abstract

Cardiovascular diseases continue to pose a significant global health problem, accounting for approximately 32% of all global deaths according to the World Health Organization. Emphasizing the urgent need for efficient and reliable diagnostic methods, electrocardiogram (ECG) analysis offer critical informations on cardiac state and activity, making it fundamental for identifying a wide range of heart conditions. This paper presents an enhanced approach for ECG signal classification using deep convolutional neural networks (CNNs). We introduce several improvements to conventional CNN architectures, including statistical resampling, targeted data augmentation, residual connections, and attention mechanisms to address the significant class imbalance present in ECG datasets. Using the MIT-BIH Arrhythmia Database, our model classifies five arrhythmia categories: Normal, Supraventricular, Ventricular, Fusion, and Unknown beats, achieving 98% accuracy. Our experimental results, including a comprehensive ablation study, demonstrate that each component of our architecture contributes to the overall performance, with data augmentation and class weighting having the most significant impact. The enhanced model shows particular strength in handling the severe class imbalance (82.8% Normal beats) common in real-world ECG datasets, making it suitable for clinical applications. Comparative analysis with recent state-of-the-art methods reveals that our approach offers a favourable balance between classification accuracy and computational efficiency.