Advanced Cardiac Signal Processing with Federated Learning and IoT for Remote Health Monitoring

Authors:

Abstract
Cardiovascular diseases are now ranked among the leading causes of death; medical care requires efficient, immediate cardiac monitoring devices. This study provides an original approach incorporating Federated Learning (FL) with IoT-funded wearable devices for raw cardiac signal processing in real-time, aiming to avoid privacy problems and reduce time delay associated with centralized approaches. In this study, the performance of the developed system was evaluated using the two ECG databases that are freely available in the public domain, the MIT-BIH Arrhythmia Database and the PhysioNet Challenge Database. This proposed model achieved a performance of 97.8% in detecting cardiac disorders such as arrhythmias and atrial fibrillations. However, measuring by accuracy, it scored 96.5% overall accuracy while concerning the recall, it scored 98.2%, which accounted for an F1 measure of 97.3% overall score. By minimizing latency, the architecture offered a 35% improvement over the traditional centralized CM systems, which makes it very suitable for real-time monitoring. By applying differential privacy methods, the privacy loss parameter was set to ε = 1.5, thus ensuring the privacy of sensitive patient data. The study demonstrates how the proposed system provides a privacy-preserving, real-time, and scalable approach to monitoring heart health, making it a valuable contribution to biomedical signal processing. Current work shall focus on expanding the above model to other biological signals and improving its performance for more widespread use in clinical medicine.