摘要
提出一种基于集成算法的联邦学习模型(Stacked Federated Learning,Stacked-FL)。模型将集成算法与联邦学习相结合,基于RSA加密与同态加密的隐私保护方法,实现多方数据去中心化处理,并设计了数据预处理、样本对齐、联邦集成算法模块。实验结果表明,该模型在预测准确性方面有显著提升,相比传统的数据中心化模型,其安全性得到了保证。实验使用Lending Club贷款数据集,选取MAE与RMSE作为评价指标,根据Stacked-FL算法的结果显示,其性能相比SecureBoost算法分别提升7.98%和0.83%。在保障用户隐私安全的同时,也提高了预测精确度,相比传统的联邦算法泛化能力增强,该模型满足中小型敏感数据集联合建模的基本需求,为中小型银行定制化信贷提供了更有效的解决方案。
In this paper,a federation learning model(Stacked Federated Learning,Stacked-FL)based on an integrated algorithm is proposed to try to solve the above problem.Data preprocessing,sample alignment,and federated ensemble algorithm modules are built using ensemble algorithms and federated learning,based on RSA encryption and homomorphic encryption for privacy protection.The experimental results show that the model has significant improvement in data security and prediction accuracy.The experiments use the Lending Club loan dataset and MAE and RMSE are selected as evaluation metrics,and the results based on the Stacked-FL algorithm show that its performance is improved by 7.35%and 0.39%,respectively,compared to the SecureBoost algorithm.The model meets the basic needs of joint modeling of small and medium-sized sensitive datasets and provides a more effective solution for customized credit for small and medium-sized banks by safeguarding user privacy and security while also improving prediction accuracy and enhancing generalization capability compared to traditional federated algorithms.
作者
岳靖轩
陈志雨
刘钢
YUE Jingxuan;CHEN Zhiyu;LIU Gang(School of Computer Science&Engineering,Changchun University of Technology,Changchun 130102,China)
出处
《长春工业大学学报》
2023年第4期313-322,共10页
Journal of Changchun University of Technology
基金
吉林省省级产业创新专项资金计划项目(2017C034-4)。
关键词
联邦学习
集成算法
贷款定价
同态加密
数据安全
federated learning
ensemble algorithm
loan pricing
homomorphic encryption
data security.
