Privacy-Preserving Collaborative Deep Learning for Cyber Threat Identification in 5G Healthcare 4.0
Arun Kumar Dey, Govind P. Gupta, Satya Prakash Sahu · 2024
Healthcare 4.0, the fourth revolution in the healthcare industry, drives rapid technological advancements and enhances the quality of patient care. Healthcare 4.0 data, sourced from intelligent networks like 5G, is rapidly growing in volume but raises significant security and privacy concerns due to its vulnerability to cyber threats and data storage environment. To address this issue, collaborative deep learning was introduced to train and detect cyber threats without sharing private data. However, many literatures indicates that privacy can still be compromised during the training process in collaborative deep learning. In addition, the training cost of collaborative deep learning is also relatively high. In this study, a privacy-preserving collaborative deep learning framework, named PPCDL, is suggested to identify and prevent cyber threats in Healthcare 4.0 ecosystems. In this framework, on the peer side, Paillier-based homomorphic encryption is utilized to encode the local training model shared by the peers to protect the privacy and security of data. Moreover, a MobileNet-V1-driven lightweight deep learning model that reduces training costs is used to identify cyber threats. Comprehensive experiments on a realistic 5G-NIDD dataset reveal that the proposed PPCDL framework excels in identifying multiclass cyber threats to healthcare systems with a precision of 99.75% compared to state-of-the-art frameworks.