Hybrid Machine Learning Framework with Data Analytics Model for Privacy-Preserved Intelligent Predictive Maintenance in Healthcare IoT

Abstract

Federated Learning (FL) is a cutting-edge approach for developing machine learning (ML) models using distributed datasets while preserving data privacy and ownership. FL is particularly suited for Internet of Things (IoT) networks due to its decentralized nature, which supports In-Edge AI and maintains data locality. However, FL's complexity poses challenges in analyzing system health, making it crucial to develop robust strategies for monitoring and evaluation. This research introduces a hybrid machine learning architecture that combines FL with the Adaptive Moving Window Regression (AMWR) technique. Specifically, we employ Federated Learning with Dynamic Regularization (FedDyn), where model architecture and training configurations are established centrally and disseminated to clients, who contribute to the model while ensuring differential privacy. This approach termed Federated Learning with Dynamic Regularized Adaptive Moving Window Regression (FedDyn AMWR), demonstrates significant improvements in system reliability, availability, maintainability, and safety. Experimental comparisons with existing methods show that FedDyn AMWR offers substantial advantages in accuracy, computational efficiency, and security, making it a promising solution for complex multi-object systems' health management and maintenance strategies in IoT-based healthcare.