A Traceable Four-party Privacy-preserving Machine Learning Framework with Application to Vehicular Networks

Abstract

Privacy-preserving machine learning has become a key technology for protecting sensitive user data and has been widely applied in fields such as vehicular networks, cloud computing, and the Internet of Things (IoT). However, model training often relies on large-scale datasets, and thus achieving efficient computation while ensuring data privacy remains a core challenge. Taking vehicular networks as an example, vehicle security operation centers (VSOCs) operated by different vendors need to collaborate in analyzing security threats, but due to commercial competition and data sensitivity, raw data cannot be directly shared. To address this challenge, we propose a privacy-preserving machine learning framework that supports the tracing of malicious participants. The design, based on the honest majority assumption, implements a four-party interactive protocol using secret sharing, without the need for a trusted third party to guarantee output delivery. Additionally, when data exchange results in conflicts, the protocol can quickly identify the malicious participant. Thus, it provides efficient mechanisms for malicious participant tracing, significantly enhancing system robustness and real-world deployability. Furthermore, we optimized the robust multiplication protocol by reducing the required number of ring elements to 7, leading to a 12.5%reduction in communication overhead. Our protocol enables collaborative multiparty analysis of vehicular data across competing manufacturers with diverse datasets, achieving information-theoretic privacy while overcoming data silos and preserving data confidentiality. Comparative experiments using logistic regression and neural networks on multiple datasets (MNIST, CICIDS2017, UNSW-NB15) demonstrate that our approach, while providing stronger security, achieves an average efficiency improvement of 1.38 times compared to an existing four-party protocol (Tetrad).

Keywords

Privacy Preserving Machine Learning / Secure Multi-party Computation / Malicious User Identity Tracing / Guaranteed Output Delivery / Vehicular Networks

RIGHTS & PERMISSIONS

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.