S-CreditRisk: A Decentralized and Privacy-Preserving Credit Risk Assessment Framework Using Federated Learning

This paper proposes the S-CreditRisk framework, an innovative solution for credit risk assessment that uses federated learning to allow financial institutions to jointly build models without risking data privacy or violating regulatory requirements. Unlike conventional machine learning and deep learning methods that require centralized data storage-posing privacy and scalability challenges, whereas S-CreditRisk enables distributed data processing, allowing institutions to use varied datasets without direct data sharing. The proposed framework enhances both the accuracy and quality of credit risk evaluations. Furthermore, the incorporation of blockchain technology with off-chain storage mechanism, i.e., InterPlanetary File System (IPFS), guarantees that model updates are transparent and traceable, promoting trust among the collaborating institutions. In the proposed framework, accuracy is obtained as <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 2. 6 \%}$</tex> and it shows improvements in ROC-AUC and recall, which demonstrates its effectiveness compared to the existing work. Overall, S-CreditRisk offers a scalable, efficient, secure, and privacy-focused method for managing credit risk in financial services.