Enhanced Multi-Class Arrhythmia Detection Using Generative Adversarial Networks for Minority Class Augmentation

Author First name, Last name, Institution

Heba Ismail, Zayed University
Mohamed Adel Serhani, University of Sharjah
Benjamin C.M. Fung, Université McGill

Document Type

Article

Source of Publication

Cognitive Computation

Publication Date

12-27-2025

Abstract

Arrhythmia is a prevalent cardiac disorder and a growing public health concern. Electrocardiogram signals aid in early diagnosis and prediction of Arrhythmia. Early detection of rare arrhythmias, such as atrial fibrillation, is critical for timely intervention and mortality reduction. However, the predictive accuracy for Arrhythmia minority classes remains limited due to extreme imbalance in training datasets. To address the imbalance of Electrocardiogram datasets, this paper proposes an enhanced Generative Adversarial Network-based approach for augmenting Electrocardiogram datasets to improve predictive accuracy for minority classes. The proposed method utilizes an enhanced Generative Adversarial Network model to synthesize representative Electrocardiogram signals, specifically targeting the inclusion of rare arrhythmia classes. The augmented dataset is then passed to a predictive model for multi-class arrhythmia detection. The proposed method is validated using three publicly available benchmark datasets: UCI Machine Learning-Arrhythmia, PhysioNet CINC 2017, and MIT-BIH Arrhythmia datasets. The results of the data synthesis module demonstrate a highly accurate generation of Electrocardiogram signal data samples including rare Arrhythmia classes. Moreover, the classification results show that the proposed method achieves high predictive accuracy, especially for minority classes. The proposed method achieved overall accuracies of 98.098%, 95.488%, and 99.880% for the UCI, CINC, and MIT-BIH datasets, respectively. The results of the proposed model outperformed state-of-the-art predictive accuracies for Arrhythmia minority class prediction.

DOI Link

10.1007/s12559-025-10520-3

ISSN

1866-9956

Publisher

Springer Science and Business Media LLC

Volume

18

Issue

1

Disciplines

Computer Sciences | Medicine and Health Sciences

Keywords

Arrhythmia, Data synthesis, Deep learning, Generative adversarial networks, Minority classes

Scopus ID

105026240265

Recommended Citation

Ismail, Heba; Serhani, Mohamed Adel; and Fung, Benjamin C.M., "Enhanced Multi-Class Arrhythmia Detection Using Generative Adversarial Networks for Minority Class Augmentation" (2025). All Works. 7744.
https://zuscholars.zu.ac.ae/works/7744

Indexed in Scopus

yes

Open Access

no