Privacy-Preserving Credit Scoring Using CKKS-Based Fully Homomorphic Encryption

초록

Credit scoring plays a crucial role in financial decision-making but inherently relies on highly sensitive personal and financial data, raising significant privacy concerns. Conventional credit scoring systems expose raw data to centralized servers, creating vulnerabilities to misuse, unauthorized access, or data breaches. To address this gap, we propose a privacypreserving credit scoring framework based on the CKKS scheme of Fully Homomorphic Encryption (FHE), which enables arithmetic operations to be performed directly on encrypted data without decryption. Our approach allows clients to encrypt their financial attributes before submission, while the credit scoring model executes inference entirely in the encrypted domain, ensuring that sensitive inputs remain confidential throughout the process. The results demonstrate that encrypted inference achieves accuracy comparable to plaintext computation, with negligible loss due to CKKS approximation, while incurring overhead in terms of latency, ciphertext expansion, and memory usage. We further analyze the trade-offs between performance and security levels to identify practical parameter settings. This work represents a novel demonstration of end-to-end encrypted credit scoring using CKKS-based FHE, highlighting its potential for deployment in privacy-sensitive financial applications. © 2025 IEEE.

키워드