Quantifying Privacy-Utility Trade-Offs in Cross-Bank Fraud Detection via Differentially Private Federated Learning

This paper quantifies the privacy-utility trade-off in cross-bank fraud detection using Differentially Private Federated Learning. Our empirical findings show that while strict privacy budgets hamper minority class detection in highly imbalanced datasets, an optimal parameter tuning (ε ≈ 4.0, C ≈ 1.0) achieves near-baseline detection capabilities while satisfying rigorous differential privacy standards. Future work will investigate adaptive gradient clipping and secure multi-party computation (SMPC) hybrid architectures to further reduce required noise levels.

·     . Wei, J. Li, M. Ding, C. Ma, H. H. Yang, F. Farokhi, S. Jin, T. Q. S. Quek, and H. V. Poor, "Performance Analysis and Optimization of Differentially Private Federated Learning," IEEE Transactions on Information Forensics and Security, vol. 15, pp. 3098–3112, 2020. (Key paper for mathematically quantifying the privacy-utility tradeoff)   Q. Yang, Y. Liu, T. Chen, and Y. Tong, "Federated Machine Learning: Concept and Applications," ACM Transactions on Intelligent Systems and Technology (TIST), vol. 10, no. 2, pp. 1–19, 2019. (Foundational paper for Cross-Silo Federated Learning in Banking and Finance)   M. Abadi, A. Chu, I. Goodfellow, H. B. McMahan, I. Mironov, K. Talwar, and L. Zhang, "Deep Learning with Differential Privacy," in Proc. ACM SIGSAC Conference on Computer and Communications Security (CCS), 2016, pp. 308–318. (Foundational paper for Differential Privacy budget $\epsilon, \delta$ noise addition)   W. Zheng, L. Yan, C. Gou, and F.-Y. Wang, "A Privacy-Preserving Federated Learning Framework for Financial Fraud Detection," in Proc. IEEE International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom), 2020, pp. 1432–1439. (Direct application to Financial Fraud Detection)   ·      R. C. Geyer, T. Klein, and M. Nabi, "Differentially Private Federated Learning: A Client Level Perspective," arXiv preprint arXiv:1712.07557, 2017. (Focuses on client-level/bank-level differential privacy guarantees)