A Blockchain-federated Learning Approach to Fraud in Banking Sector
Abstract
Fraud continues to be a universal issue for financial institutions. The need for compliance
leads to the biggest pain point that stands in the way of widespread adoption by research:
poor-quality training data. This makes it even harder for banks to share their data openly, and
organisations will always have concerns with data privacy and the potential exposure of
proprietary information. In order to protect data privacy for deep learning algorithms,
federated learning (FL) has become popular. By locally retaining client data, FL's application
to fraud reduces privacy risks but tackles only part of the problem; it still faces barriers at
each stage in such real-world deployments: such as centralised coordination failures, lack of
adequate security, scarcity of quality and relevant data due to data protection laws, and
regulatory compliance challenges. This work proposes implementing a Blockchain Federated
Learning (BCFL) methodology that integrates into this space addressing the issues raised.
The BCFL fraud detection artefact is developed collectively from multiple clients while
providing governance for a synergistic model training process in the banking sector. This
paper provides insights into how BCFL can solve the problem of complex data dynamics
while also ensuring privacy and security. It can use different sets of heterogeneous data from
different sources, which addresses the problem of a lack of individual systems, nudging for
collaboration at a large scale. This approach leverages diverse datasets from multiple sources,
addressing the issue of limited data availability within individual systems, showing potential
in financial data-sharing frameworks for augmenting privacy and security through BCFL. To
validate the feasibility of the application, Chapter 4 conducts an empirical comparison and
analysis. It also discusses how BCFL plays a role in the banking sector and what unimagined
endeavours are waiting for this innovation to reform both sensitive data privacy and security.
This study used a simulated dataset, which might not accurately capture all the nuances of
real fraud data over time. On the whole, BCFL is an entire blueprint designed to speed up
innovation in financial technology and enhance trust as well as resilience. This indicates an
application for collaborative training of a central model that can be used by banks with small
datasets to build their fraud detection approach.
Publisher
University of Nairobi
Rights
Attribution-NonCommercial-NoDerivs 3.0 United StatesUsage Rights
http://creativecommons.org/licenses/by-nc-nd/3.0/us/Collections
The following license files are associated with this item: