Review of Hybrid Deep Learning Techniques for Robust ECG Signal Classification by Addressing Noise and Class Imbalance

This review paper presents a comprehensive analysis of recent advancements in hybrid deep learning techniques for automated electrocardiogram (ECG) signal classification, with a central focus on overcoming challenges such as signal noise and class imbalance. Given the pivotal role of ECG analysis in the early detection of cardiovascular diseases, there is a growing demand for intelligent, robust, and clinically deployable systems. The paper examines hybrid model architectures that integrate Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and attention mechanisms to effectively capture both spatial and temporal features of ECG signals. Key studies are reviewed in terms of their preprocessing strategies, feature optimization methods, and noise mitigation techniques. In addition to architectural insights, the paper outlines a structured taxonomy of hybrid approaches based on model composition, preprocessing methods, imbalance handling techniques, evaluation metrics, and deployment environments. It also explores emerging application areas, including real-time monitoring systems, telecardiology platforms, and clinical decision support tools. Common research limitations—such as computational complexity, limited interpretability, and real-world generalizability—are critically discussed. The review concludes with recommendations to guide future work toward scalable, explainable, and clinically relevant deep learning frameworks, ultimately aiming to support the next generation of intelligent cardiovascular healthcare systems.