Privacy-Preserving Collaborative Information Sharing through Federated Learning–A Case of the Insurance Industry

Dong, Panyi; Feng, Runhuan; Quan, Zhiyu; Wang, Tianyang

Permalink

https://hdl.handle.net/2142/121811 Copy

Description

Title

Privacy-Preserving Collaborative Information Sharing through Federated Learning–A Case of the Insurance Industry

Author(s)

• Dong, Panyi
• Feng, Runhuan
• Quan, Zhiyu
• Wang, Tianyang

Issue Date

2023

Keyword(s)

• Federated learning
• Privacy preservation
• Insurance loss modeling
• Risk assessment

Abstract

The notable successes brought about by Machine Learning collaborations are reshaping our society, which, despite their promising potential, can be hindered by privacy concerns regarding the practice of centralized data collection shared among various industries that retain confidential information, particularly in the case of insurance. In this study, we propose the application of Federated Learning (FL), an emerging privacy- preserving distributed learning approach, as a solution to those crucial privacy concerns in the insurance domain and to enable real-world industrial collective learning. In particular, we utilize the innovations in FL to address two of the most pressing challenges, shortage of data volume and data variety, that the insurance industry is facing. By fostering collaborations among insurance companies or through cross-industrial part- nerships, all private ”data owners”, typically considered ”data silos” in the conventional perspective, can be connected by an aggregation server to generate a consensus model with accumulated insights without sharing raw data. Such an FL framework, where the traditional central server is substituted by the aggregation server, facilitates collaborative learning in the context of privacy-preservation and empowers a more comprehensive analysis of risk assessment, which is further supported by our empirical experiments for the task of insurance loss modeling. By reflecting the applications to real-life, the collaborations possess the potential of creating such an environment in which the policyholders/customers receive the fairer pricing, the capability of risk management of insurance companies is enhanced, and the regulators gain deeper insights into the industry.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121811

Owning Collections