Differentially Private Federated Autoencoder Framework for Securing Internet of Medical Things

As healthcare systems adopt more connected medical devices, concerns about data security and patient privacy are growing rapidly. Internet of Medical Things (IoMT) devices often deal with sensitive health information, making intrusion detection both important and tricky. In this work, we design a privacy-friendly intrusion detection framework leveraging autoencoder, federated learning and differential privacy. It starts by cleaning and compressing raw network traffic using neural autoencoders. Instead of sending all data to one place, we train models distributively on the devices using federated learning. To make sure patient information stays private, we also add Differential Privacy. Relevant experiments were carried out to test the proposed framework using CICIoMT2024 dataset. For the proposed model, the Random Forest method worked the best, achieving 99.39% accuracy without privacy protection and 98.34% ensuring privacy. These results show that it is possible to catch threats quickly while still keeping sensitive data safe and meeting the unique needs of medical networks.