Privacy Preserving Data Analytics in 5G-Enabled IoT for the Financial Industry

TL;DR

This paper presents a privacy-preserving architecture using homomorphic encryption for secure, fast credit assessment in 5G-enabled IoT financial services, enabling encrypted data processing on edge servers.

Contribution

It introduces a novel architecture and protocol leveraging homomorphic encryption for secure, real-time credit assessment on edge servers in the financial industry.

Findings

Achieves short response times suitable for real-time applications

Maintains reasonable prediction accuracy with encrypted data

Demonstrates feasibility of privacy-preserving edge computing in finance

Abstract

Next-generation wireless networks like 5G promise faster speed, shorter latency, and the ability to connect more devices. Such benefits are set to make drastic changes to the future society, empowering smart cities, enabling autonomous cars, enhancing business processes, changing consumer behaviors, etc. In the financial industry, banks evaluate the deployment of Internet of Things (IoT) technologies and edge computing for better customer engagement, e.g., mobile branches on a vehicle, micro-ATM, self-service digital panel, etc. One of the trends is breaking down monolithic business application systems into micro-services for deployment on distributed edge servers, thus reducing network latency and improving services. Such movements pose challenges in protecting the security and privacy of business data between access points. This paper introduces a new architecture and protocol to tackle a use case for the financial industry. The solution assumes deploying a credit assessment model on an edge server. The model accepts and processes encrypted data submitted by potential customers seeking online credit assessments. The encrypted assessment results are sent back to the customers for decryption and interpretation. The data transmission rides on asynchronous communication, and the data protection uses Homomorphic Encryption. A proof-of-concept experiment shows that the proposed method can be achieved with a short response time and a reasonable prediction accuracy.