DEEP ENCODING THE HEART: AUTOENCODER-DRIVEN ECG ANOMALY RECOGNITION

Abstract

Electrocardiogram (ECG) analysis plays a vital role in diagnosing cardiovascular diseases, which remain the leading cause of mortality worldwide. Traditional methods for ECG anomaly detection often rely on handcrafted features and supervised learning, which require large labeled datasets and may fail to capture subtle irregularities. To address these challenges, this study proposes an autoencoder-driven deep learning framework for unsupervised detection of ECG anomalies. The model leverages the autoencoder’s ability to learn compressed representations of normal cardiac rhythms and identifies anomalies based on reconstruction errors. By training exclusively on normal ECG signals, the system efficiently distinguishes between healthy and abnormal patterns without prior knowledge of specific anomaly types. Experimental results on publicly available ECG datasets demonstrate high sensitivity and specificity, outperforming conventional machine learning approaches. The proposed framework not only enhances early detection of cardiac disorders such as arrhythmias but also reduces dependency on annotated datasets, making it highly scalable for realworld clinical applications. This work highlights the potential of deep autoencoders as a powerful tool for advancing automated, accurate, and timely cardiac health monitoring.