Edge-Optimized Federated Learning Using Differential Privacy for Secure Patient Monitoring in AI-Driven MIoT Healthcare Systems

Abstract

Real-time surveillance, predictive diagnosis, informed judgement are changing healthcare of AI and
MIoT. Technology-blend models contribute to this development. Nevertheless, the IoT devices produce
high sensitive patient data that implicates privacy, security and scalability issues. To provide an AI
safeguards and effective patient monitoring mechanism in burgeoning MIoT AICT, we introduce edge-
centric federated learning with differential privacy for the purpose of realistic cost-effective safe patient
monitoring. Patient entries are saved to (and reside with at) the network edge in built-in bedside or IoT
devices. In the mean time, in cooperative nodes parametersized are learned by training of non-raw-data
transmitting models through the proposed system. Unsurprisingly, differential privacy affords shared
model updates tuned noise. Edge computing improves the quality of remote patient care due to limited
bandwidth requirements, supports always-on/low-power operation and ensures low latency. The proposed
method is tested on real healthcare datasets in terms of accuracy, privacy and computational efficiency.
These results indicate that edge-optimized federated learning can actually improve prediction and achieve
communication efficiency at the same time as it fights privacy concerns. The goal of this project is to have
a permanent and scalable solution for patient privacy in future healthcare systems without compromising
the monitoring safety. The proposed approach paves the way for federated AI models in IoT.