Empirical Evaluation Model for Differential Privacy in Federated Learning for Cyber Threat Intelligence Sharing
Keywords:
Cyber threat intelligence, Differential privacy, Federated learning, Heterogeneous differential privacy federated learning (HetDP-FL), Member inference attackAbstract
Today, the threat landscape is evolving rapidly, with many organizations continually facing complex and malicious cyber threats. Collective Cyber Threat Intelligence (CTI) sharing has become a proactive approach for addressing cyber threats. However, there is a drawback in sharing raw CTI data due to privacy and confidentiality concerns. Although Federated Learning (FL) remains vulnerable to privacy attacks, Differential Privacy (DP) has become the de facto standard for privacy. However, there is a critical gap in how DP affects utility and the extent of privacy loss in practice. In this study, a four-layer FL model that partitions the class-imbalanced CICIDS-2017 dataset across five simulated organizations was designed. Local parameters of 3-layer Multilayer Perceptron classifiers are aggregated over ten communication rounds using a custom weight-persisting strategy; the utility of the global model was then evaluated using Macro-F1, and its empirical privacy was audited using black-box loss-threshold attacks, shadow classifiers, and reference-model Likelihood Ratio Audits (LiRA). A Heterogeneous Differential Privacy Federated Learning mechanism (HetDP-FL) that proposes and validates the selection of a client-level differential privacy budget based on the risk tolerance of each participating organization was implemented. The results of HetDP-FL improve the Macro-F1 score by +7.3% when compared to the uniform privacy budget. Results show that the benefits of federated CTI systems can be achieved by integrating formal DP guarantees with empirical privacy auditing for a particular deployment and at participant-level privacy heterogeneity.
References
T. D. Wagner, K. Mahbub, E. Palomar, and A. E. Abdallah, “Cyber threat intelligence sharing: Survey and research directions,” Computers & Security, vol. 87, Nov. 2019.
P. Alaeifar, S. Pal, Z. Jadidi, M. Hussain, and E. Foo, “Current approaches and future directions for cyber threat intelligence sharing: A survey,” Journal of Information Security and Applications, vol. 83, Jun. 2024.
A. Rejoee and J. Hannah, “Federated explainable AI system for privacy-preserving cyber threat detection and secure intelligence sharing,” International Journal of Science and Research (IJSR), vol. 15, no. 3, pp. 729–733, Mar. 2026.
A. El Ouadrhiri and A. Abdelhadi, “Differential privacy for deep and federated learning: A survey,” IEEE Access, vol. 10, pp. 22359–22380, 2022.
J. Zhao et al., “The federation strikes back: A survey of federated learning privacy attacks, defenses, applications, and policy landscape,” ACM Computing Surveys, vol. 57, no. 9, pp. 1–37, Apr. 2025.
M. Khavkin and E. Toch, “Differential privacy configurations in the real world: A comparative analysis,” in IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 11, pp. 6334–6351, Nov. 2025.
N. Aldaghri, H. Mahdavifar, and A. Beirami, “Federated learning with heterogeneous differential privacy,” arXiv, 2023.
O. I. Khalaf et al., “Federated learning with hybrid differential privacy for secure and reliable cross‐IoT platform knowledge sharing,” Security and Privacy, vol. 7, no. 3, Feb. 2024.
R. Xu, Y.-J. A. Zhang, and J. Huang, “Tackling privacy heterogeneity in differentially private federated learning,” arXiv, Feb. 2026.
N. Carlini, S. Chien, M. Nasr, S. Song, A. Terzis, and F. Tramer, “Membership inference attacks from first principles,” 2022 IEEE Symposium on Security and Privacy (SP), San Francisco, CA, USA, 2022, pp. 1897–1914.
R. Shokri, M. Stronati, C. Song and V. Shmatikov, “Membership inference attacks against machine learning models,” 2017 IEEE Symposium on Security and Privacy (SP), San Jose, CA, USA, 2017, pp. 3–18.
J. Rajamäki et al., “Implications of GDPR and NIS2 for cyber threat intelligence exchange in hospitals,” WSEAS Transactions on Computers, vol. 23, pp. 1–11, Apr. 2024.
