A privacy-preserving distributed credit evaluation model based on blockchain (2026).

Multi-source data fusion is crucial for improving the predictive performance of credit evaluation models. However, cross-organizational collaborative modeling faces challenges such as data privacy breaches, difficulty in verifying computational results, and a lack of transparency and auditability in the collaboration process. To address collaborative logistic regression modeling with vertically distributed data, this paper proposes a privacy-preserving method based on matrix transformation and further designs a trusted collaborative modeling mechanism integrating blockchain and trusted execution environment technologies. The results indicate that the proposed privacy-preserving method achieves lossless model accuracy while safeguarding data privacy and resisting collusion attacks. This method is more efficient for training and requires less communication than most federated learning schemes. The designed decentralized collaborative modeling mechanism provides a trust foundation for multiple parties without relying on a centralized trusted third party, ensuring the reliability of model results and enabling traceability and auditability throughout the entire collaborative modeling process.