Leveraging Federated Learning for Unsecured Loan Risk Assessment on Decentralized Finance Lending Platforms

Abstract

This study proposes a novel privacy-preserving unsecured loan risk assessment system that allows decentralized finance (DeFi) lending platforms to offer loans without collateral. This system leverages federated learning methods to train risk assessment models using both off-chain and on-chain data sources, to more accurately evaluate borrower default risk for unsecured loans. Moreover, this system is built on a trusted execution environment (TEE) with program-level isolation, which provides a secure and efficient solution for DeFi platforms to offer unsecured loans. The effectiveness of this platform is validated through a set of simulation experiments. These experiments underscore the capability of the federated learning models to accurately assess borrower default risk while preserving stringent data privacy standards. The unique and innovative system design we proposed offers significant advancements for DeFi lending platforms. These improvements have the potential to greatly enhance DeFi platforms’ inclusiveness by offering unsecured loans while maintaining efficiency, and security.