Intrusion Detection System for IoMT through Blockchain-based Federated Learning

Abstract

Federated Learning (FL) is a feasible technology to collaboratively train a model without sharing private data. This approach differs from traditional machine learning techniques, which aggregate local datasets in a single server. Thus, FL is adopted in the Healthcare sector to preserve the privacy of collected sensitive medical data from heterogenous and resource-constrained Internet of Medical Things (IoMT). However, FL requires aggregating all trained local models in a central server that presents a single point of failure. To address this issue, we propose a novel Blockchain-based Federated Learning architecture, which is applied to detect malicious network traffic in IoMT environments. In this paper, the proposed architecture takes advantage of a Hyperledger Fabric channel coupled with FL to manage efficiently and securely the learning process of an Intrusion Detection System (IDS). The Blockchain channel replaces the commonly used central server with the traditional FL approach. Moreover, the proposed approach benefits from the inherent features of Blockchain and FL. Besides, it secures patient data collection from the IoMT by examining the network traffic for unauthorised behaviour or policy breaches.