Intrusion Detection System for IoHT Devices using Federated Learning
Fatemeh Mosaiyebzadeh, Seyedamin Pouriyeh, Reza M. Parizi, Meng Han, Daniel Macêdo Batista
Abstract
With the growing number of sensitive data transmit-ted in IT infrastructures, healthcare organizations and compa-nies that generate users' wearable data have become a target for attackers. To protect electronic healthcare data, Internet of Healthcare Things (IoHT) devices must be protected by robust Intrusion Detection Systems (IDS) to provide a secure environment. Since it is undesirable to collect this data and perform machine learning tasks directly, recently, to preserve users' privacy, federated learning has obtained attention from the government and healthcare organizations. Unlike the cen-tralized paradigm, federated learning is a privacy-aware machine learning framework designed to analyze data without sharing it. This paper proposes a deep neural network in federated learning (DNN-FL) to detect anomalies in the IoHT traffic that may result in security threats. We evaluate the detection performance of our proposal using metrics such as accuracy and precision. The proposed DNN-FL is validated using the wustl-ehms-2020 and ECU- IoHT datasets. It reached 91.40% of accuracy in detecting attacks in the first dataset and 98.47% in the second. All the developed source code in this work is being made publicly available to ensure reproducibility.