Perfumer Optimization - Enhanced Graph Neural Networks for Efficient and Accurate COVID-19 Risk Prediction in IoT/IoMT Healthcare

A great imperative to develop healthcare monitoring systems that are real-time, accurate and lowlatency has emerged since the COVID-19 pandemic. Traditional cloud-based solutions and machine learning methods face problems of latency, poor flexibility and inferior feature representation which hinder significantly their utilization in IoT- and IoMT-powered healthcare systems. To solve these problems, a Perfumer Optimization- Attentionbased Dynamic Multilayer Graph Neural Network (POADMGNN) for COVID-19 risk prediction over fog-assisted healthcare systems has been presented in this research. The platform provides secure citizen registration, ongoing data collection from wearables and the environment through IoT sensors, and effective preprocessing at the fog layer, in which the Superb Fairy-Wren Optimization Algorithm makes choice of the most important features. The features are then differentiated via ADMGNN which is designed based on multilayer graph, temporal dynamics and attention mechanism to capture complicated health issues. In addition, the Perfumer Optimization Algorithm adjusts hyperparameters to improve model performance, which includes real-time decision support and alert modules for instant notification, personalized guidance and facility navigation as well as synchronized storage which translates to quick local data access, and secure cloud-based epidemic surveillance. Experiment results prove that PO-ADMGNN achieves better performance with almost 98% classification precision, an F measure of 0.978 for the negative class, a $7.6 \times$ response time reduction (380 ms in the cloud vs. 50 ms in the fog), and a 54% latency reduction with an optimum peak accuracy of 0.9777. These results validate that PO-ADMGNN provides accurate, adaptive, and efficient healthcare monitoring with effective COVID-19 control.