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

• DOI: 10.48550/arxiv.2402.14983
• Publication Date: 2024-02-22
• Authors: Panyi Dong; Zhiyu Quan; Brandon Edwards; Shih-han Wang; Runhuan Feng; Tianyang Wang; Patrick Foley; Prashant Shah
• Keywords: Insurance industry; Information Sharing

The report demonstrates the benefits (in terms of improved claims loss modeling) harnessing value Federated Learning (FL) to learn a single model across multiple insurance industry datasets without requiring themselves be shared from one company another. application FL addresses two most pressing concerns: limited data volume and variety, which are caused by privacy concerns, rarity claim events, lack informative rating factors, etc.. During each round FL, collaborators compute improvements on using their local private data, these insights combined update global model. Such aggregation allows for an increase effectiveness in forecasting losses compared models individually trained at collaborator. Critically, this approach enables machine learning collaboration need raw leave infrastructure respective owner. Additionally, open-source framework, OpenFL, that is used our experiments designed so it can run confidential computing as well with additional algorithmic protections against leakage information via updates. In such way, implemented privacy-enhancing collaborative technique challenges posed sensitivity traditional solutions. This paper's also expanded other areas including fraud detection, catastrophe modeling, etc., have similar incorporate into collaborations. Our framework empirical results provide foundation future collaborations among insurers, regulators, academic researchers, InsurTech experts.