Edge-Driven Machine Learning for Real-Time Health Anomaly Detection in IoT-Based Healthcare: A Modular Framework

Abstract

The rapidly expanding field of healthcare IoT has the potential to transform patient monitoring and intervention. Real-time anomaly detection is necessary to safeguard patient information and device integration in light of these new security threats brought about by this interconnectedness. Integrating the Internet of Things (IoT) into healthcare enhances patient care while also improving the performance and reliability of healthcare delivery system. Ethical concerns, interoperability, and data security must be resolved if the benefits of IoT in healthcare are to be fully realized. Patients' health data is measured by IoT devices on a regular basis and shared with a server for additional analysis. Various machine learning (ML) approaches are employed on the server to help with early disease identification and to send out alerts when vital signs deviate from normal. IoT devices regularly measure patient health data and send it to a server for further analysis. The server employs a variety of ML techniques to assist in the early detection of diseases and to notify users when vital signs diverge from normal. Patient health data is routinely measured by IoT devices and sent to a server for additional analysis. The server uses a range of edge-driven ML models to alert users when vital indicators deviate from normal to anomaly identification. In this study, we introduce an edge-driven ML for real-time health anomaly detection in IoT-based healthcare: a modular framework (EDML-RHADIOTF). The EDML-RHADIOTF involves data preprocessing, feature extraction and anomaly classification. In comparison to previous solutions, the suggested EDML-RHADIOTF model shown an outstanding enhancement in anomaly detection based on an analysis of its performance outcomes. With a 96.55% accuracy rate in both reduced and full feature space, SVM performed better across anomaly classification of IoT attacks. A reduction in computing response time, which is necessary for quick response and real-time attack detection, complemented this improvement.