Scholarlink

<p>In today's data-driven banking industry, institutions rely heavily on data for decision-making, particularly in loan approvals. However, sharing sensitive customer data across banks presents significant privacy risks, necessitating secure and privacy-preserving solutions. This paper proposes a privacy-preserving, multi-institutional loan approval framework based on Federated Learning (FL), explicitly utilizing the Federated Flower framework. FL enables banks to train machine learning models collaboratively without sharing raw data, ensuring that customer information remains secure while benefiting from collective model improvements. We evaluate various FL methods within the Federated Flower framework, including FedAvg, FedCodl, FedProx, and FedKT, which offer flexibility and scalability to address challenges such as system heterogeneity, data imbalance, and knowledge transfer. The framework ensures robust privacy preservation while achieving high predictive performance across multiple institutions. The proposed model is tested on loan prediction datasets such as Loan Data, HMEQ, and Taiwan. Among the different FL methods, FedKT outperforms the others, achieving performance similar to centralized models while maintaining privacy. Additionally, the performance of FL is compared with traditional machine learning techniques, such as Logistic Regression and Decision Trees, with FL demonstrating superior accuracy. The results demonstrate that the Federated Flower framework offers a secure, scalable, and privacy-preserving solution for multi-institutional loan processing, advancing predictive modelling for sensitive, distributed data environments in the banking sector.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Privacy-Preserving Bank Loan Approval with Federated Learning: A Secure and Collaborative Multi-Institutional Prediction Framework

  • 期刊论文

  • 发表日期：

    2025年5月22日

查看全文
  • Vankamamidi S. Naresh,
  • M. Thamarai,
  • Sivaranjani Reddi

摘要

In today's data-driven banking industry, institutions rely heavily on data for decision-making, particularly in loan approvals. However, sharing sensitive customer data across banks presents significant privacy risks, necessitating secure and privacy-preserving solutions. This paper proposes a privacy-preserving, multi-institutional loan approval framework based on Federated Learning (FL), explicitly utilizing the Federated Flower framework. FL enables banks to train machine learning models collaboratively without sharing raw data, ensuring that customer information remains secure while benefiting from collective model improvements. We evaluate various FL methods within the Federated Flower framework, including FedAvg, FedCodl, FedProx, and FedKT, which offer flexibility and scalability to address challenges such as system heterogeneity, data imbalance, and knowledge transfer. The framework ensures robust privacy preservation while achieving high predictive performance across multiple institutions. The proposed model is tested on loan prediction datasets such as Loan Data, HMEQ, and Taiwan. Among the different FL methods, FedKT outperforms the others, achieving performance similar to centralized models while maintaining privacy. Additionally, the performance of FL is compared with traditional machine learning techniques, such as Logistic Regression and Decision Trees, with FL demonstrating superior accuracy. The results demonstrate that the Federated Flower framework offers a secure, scalable, and privacy-preserving solution for multi-institutional loan processing, advancing predictive modelling for sensitive, distributed data environments in the banking sector.