Detecting Electrocardiogram Arrhythmia Empowered With Weighted Federated Learning

Abstract

In this study, a weighted federated learning approach is proposed for electrocardiogram (ECG) arrhythmia classification. The proposed approach considers the heterogeneity of data distribution among multiple clients in federated learning settings. The weight of each client is dynamically adjusted according to its contribution to the global model improvement. Experiments on public ECG datasets show that the proposed approach outperforms traditional federated learning and centralized learning methods in terms of accuracy and robustness. On the client side, the suggested federated learning (FL) approach had an accuracy of 0.93, sensitivity of 0.98, specificity of 0.82, miss classification rate of 0.07, precision of 0.06, FPR of 0.01, and FNR of 0.01. FL has 0.98 accuracy, 0.99 sensitivity, 0.91 specificity, 0.02 miss classification rate, 0.10 precision, 0.01, FPR, and 0.01 FNR on the server. The server-side federated learning approach outperforms the client-side in accuracy, sensitivity, specificity, miss classification rates, and precision. The results indicate that the proposed weighted federated learning approach is a promising solution for ECG arrhythmia classification in a distributed environment. In short, the proposed federated learning approach applied to ECG arrhythmia detection aims to address privacy concerns and improve accuracy, while still maintaining the centralized framework and advanced algorithmic approach.