基于联邦学习的信贷风险评估模型OA
A Federated Learning-Based Model for Credit Risk Assessment
全球互联网金融的快速发展推动着以人工智能为核心的金融科技广泛渗透至信贷风险评估领域,信贷风险评估存在高维稀疏特征建模难题和数据隐私方面的挑战.鉴于此,提出一种基于联邦学习的信贷风险评估模型——FedIN模型.该模型将特征嵌入技术与Transformer编码器相结合,以高效处理类别型与连续型混合特征,并利用自注意力机制捕捉复杂的全局特征依赖.在联邦学习框架下,各参与方无需共享原始数据,仅通过交换模型参数更新即可协同训练全局模型,有效保障了数据隐私.实验结果表明,该模型在Lending Club数据集上指标均有所提升(AUC指标提升了1.46％,F1-Score指标提升了3.27％,G-mean指标提升了3.70％),验证了此模型在信贷风险评估中的有效性.
The rapid development of global Internet finance has propelled financial technology,particularly artificial intelligence,to perme-ate extensively into the field of credit risk assessment,credit risk assessment faces challenges such as high-dimensional sparse feature model-ing and data privacy.Therefore,this paper proposes a federated learning-based model named FedIN.This model integrates feature embedding techniques with a Transformer encoder to efficiently handle mixed categorical and continuous features,leveraging the self-attention mecha-nism to capture complex global feature dependencies.Within the federated learning framework,participating parties can collaboratively train a global model by exchanging only model parameter updates without sharing raw data,thereby effectively safeguarding data privacy.Experimen-tal results on the Lending Club dataset demonstrate performance improvements across key metrics:the AUC score increased by 1.46％,the F1-Score by 3.27％,and the G-mean by 3.70％,validating the effectiveness of the proposed model in credit risk assessment.
李辉;陈清怡
广西科技大学 计算机科学与技术学院(软件学院)||广西高校智能计算与分布式信息处理重点实验室,广西 柳州 545006广西科技大学 计算机科学与技术学院(软件学院)||广西高校智能计算与分布式信息处理重点实验室,广西 柳州 545006
信息技术与安全科学
联邦学习信贷风险评估自注意力机制Transformer特征嵌入深度学习
federated learningcredit risk assessmentself-attention mechanismTransformerfeature embeddingdeep learning
《软件导刊》 2026 (3)
94-99,6
评论