Design and evaluation of an AI-based Fog–IoMT architecture for COVID-19 severity prediction

The fast transmission of COVID-19 showed the essential constraints of the traditional healthcare systems, especially in real-time tracking, early severity testing, and early clinical response to the infected people. Cloud-based solutions are usually centrally located, which means high latency and low responsiveness and cannot be used in continuous and time-sensitive health surveillance. To overcome them, the proposed study will create a smart health monitoring system based on artificial intelligence, which will use a Fog-Internet of Medical Things (Fog-IoMT) architecture to predict COVID-19 cases and classify their severity early. In the proposed system, IoMT devices and Electronic Health Records are used to capture real-time physiological, symptomatic, laboratory, and demographic data, which are processed at fog nodes to facilitate ultra-low-latency analytics. It uses a novel feature selection strategy to downsize the dimensions and remove the noisy attributes, then a two-level ensemble classification framework is applied. At Level-1, severity patterns are independently learned by the weighted random forest model and the extreme gradient boosting model, whereas at Level-2, the outputs are combined together by an Artificial Neural Network to provide the final severity prediction (negative, mild, moderate, or severe). Cross validation is used to choose the best ANN setup to get the smallest prediction error possible. The proposed system is shown to be significantly more effective than traditional machine learning models in the case of experimental evaluation on a clinically guided COVID-19 dataset, with an average accuracy of 95.17, a recall of 97.67, a precision of 92.06, and a specificity of 97.26 and a low error rate. Besides prediction, the system facilitates real-time alert creation, medical reports, and geographic notifications to the patient, caregiver, and medical authority. The findings affirm that AI-based ensemble learning combined with a Fog-IoMT architecture is a useful, scalable, and low-latency method for pandemic response and future smart healthcare systems.