EdgeShield: a robust and agile cybersecurity architecture for the internet of medical things

Abstract

We present EdgeShield, a lightweight pipeline that streamlines internet of medical things (IoMT) traffic analysis by pairing aggressive dimensionality-reduction with federated model aggregation. It employs systematic preprocessing, advanced feature selection, and robust sampling to reduce computational overhead while enhancing performance. Through feature engineering techniques such as principal component analysis (PCA), targeted feature selection, and embedding methods, EdgeShield reduces dataset dimensionality by 96%, enabling near real-time detection and prevention of cyber attacks on resource-constrained edge devices. To harden the IoMT perimeter, EdgeShield trains ten lightweight edge models in just 54s and merges their parameters into a single global clas sifier with negligible extra delay. This method requires no additional training or predictions, thus accelerating deployment. Additionally, by using a compact dataset with five top-performing features and PCA with two components, EdgeShield consistently achieves accuracy levels exceeding 99.2% for individual edge models and the consolidated global model. With a built-in continuous improvement loop, EdgeShield dynamically adapts to emerging data patterns and operational conditions, driving substantial advancements in IoMT ecosystem management. This approach delivers both rapid machine learning model deployment and robust cyber attack detection, illustrating its potential to revolutionize IoMT security and elevate healthcare data integrity