ECG Arrhythmia Detector with Custom Designed Wavelet-Based Convolutional Autoencoder for Unbalanced Data
Bildiriler > Tam Metin Bildiri
Yazarlar: Öykü Eravcı, Çağla Sarvan, Nalan Özkurt
2025 Innovations in Intelligent Systems and Applications Conference (ASYU), , , Uluslararası, Tam Metin Bildiri, ss 1-6
DOI: 10.1109/asyu67174.2025.11208336
Özet
Continuous monitoring has become increasingly important to improve the quality of life for individuals at risk of arrhythmias. In this context, the use of artificial intelligence-based approaches for arrhythmia detection is playing a crucial role in modern healthcare. In this study, we propose an ECG anomaly detection system that can be used in robust and efficient evaluation of ECG beats. Custom Wavelet-based Convolutional AutoEncoder (WCAE) model is employed as ECG anomaly detector and discriminates the normal and arrhythmic beat types by using WCAE. With the use of the wavelet transform integrated autoencoder structure, both the morphological and frequency characteristics of cardiac signals, has been captured. Also, the autoencoder demonstrates the ability to handle imbalanced datasets, a common challenge in biomedical data. By operating in anomaly detection mode, the Wavelet Convolutional Autoencoder (WCAE) achieves high performance even on imbalanced datasets, as it is trained using data from only a single class.