Privacy-Preserving Machine Learning on Financial Data: Federated Learning, Differential Privacy, and Practical Deployment Challenges in Banking

Jeevan Krishna Paruchuri

doi:10.36948/ijfmr.2024.v06i04.75352

Privacy-Preserving Machine Learning on Financial Data: Federated Learning, Differential Privacy, and Practical Deployment Challenges in Banking

Author(s)	Jeevan Krishna Paruchuri
Country	United States
Abstract	Machine learning on financial data sits at an awkward intersection: the data is among the most sensitive in any industry, the regulatory regime is among the strictest, and the business value of better models is large enough to keep pulling new ML workloads into production. This survey examines the two principal families of privacy-preserving ML federated learning (training across decentralized data without centralizing it) and differential privacy (adding mathematically calibrated noise to bound the information any individual record contributes to a model) through the lens of a practitioner deploying ML in a regulated banking environment. The work is grounded in concrete operational events, including a GDPR audit that surfaced A 2022 internal GDPR audit at the partner institution discovered 14 analysts with unauthorized access to a model's training feature store; this finding directly motivated the privacy-preserving redesign reported in this paper. We review the theoretical foundations (Dwork's differential privacy framework, McMahan's FedAvg algorithm, the DP-SGD training procedure) and report the privacy-utility trade-off observed in practice: at ε=1 (strong privacy) the model accuracy degrades by approximately 5%, while at ε=5 (moderate privacy) the loss is approximately 1%**. We discuss the operational realities that make federated learning hard in banking heterogeneous data across business units, communication overhead between geographically distributed sites, convergence challenges when client distributions diverge, and the difficulty of debugging models you cannot inspect end-to-end. We argue that the privacy mechanisms themselves work; the adoption barriers are organizational, regulatory, and operational rather than algorithmic. We close with practical guidance: where federated learning earns its complexity, where centralized training with strong access controls remains the right answer, and where differential privacy is most likely to deliver its promised guarantees without crippling model utility.
Field	Computer Applications
Published In	Volume 6, Issue 4, July-August 2024
Published On	2024-07-12
DOI	https://doi.org/10.36948/ijfmr.2024.v06i04.75352

View / Download PDF File

E-ISSN 2582-2160

doi

CrossRef DOI is assigned to each research paper published in our journal.

IJFMR DOI prefix is
10.36948/ijfmr

Downloads

Research Paper Format Copyright Permission Form and Undertaking Form Cover Page Vol 8 Isu 3 Cover Page Vol 8 Isu 2 Cover Page Vol 8 Isu 2

All research papers published on this website are licensed under Creative Commons Attribution-ShareAlike 4.0 International License, and all rights belong to their respective authors/researchers.

CC-BY-SA

About IJFMR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9687-828-838	editor@ijfmr.com

International Journal For Multidisciplinary Research

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

Privacy-Preserving Machine Learning on Financial Data: Federated Learning, Differential Privacy, and Practical Deployment Challenges in Banking

Share this