A federated machine learning approach for order-level risk prediction in Supply Chain Financing
A federated machine learning approach for order-level risk prediction in Supply Chain Financing
复制标题
供应链融资中订单级风险预测的联合机器学习方法
DOI：
10.1016/j.ijpe.2023.109095
复制
发表时间：
2023-11
影响因子：
12
通讯作者：
Lingxuan Kong;Ge Zheng;Alexandra Brintrup
中科院分区：
文献类型：
--
作者：
Lingxuan Kong;Ge Zheng;Alexandra Brintrup
供应链融资（SCF）越来越多地被用作优化供应网络现金流的有效方法。随着越来越受欢迎，一些金融机构已经开始向企业提供供应链金融。然而，最近发生的各种丑闻凸显了对所涉及风险的评估效率低下。在本文中，我们认为这是由于用于评估风险的公司级特征与给出的 SCF 之间的不匹配（这是一个特定的顺序）。然而，订单级风险评估很困难，因为公司不希望与资助者共享其数据集。此外，中小企业本身可能没有足够的数据来进行订单级别的风险评估。我们提出了联邦学习（FL）框架来克服这些问题，为订单级风险评估提供了可能性。 FL 允许集体、订单级模型训练，同时保护数据所有者的隐私。航空航天业的案例研究表明，FL 可以用来预测买家的逾期付款风险，同时将性能损失降至最低。
Supply Chain Financing (SCF) is increasingly utilised as an effective method for optimising cash flows in supply networks. With increased popularity several financial institutions have begun offering SCF to businesses. However various recent scandals have highlighted inefficiencies in the evaluation of risks involved. In this paper we argue this is due to a mismatch between the firm-level features used to evaluate risk and what SCF is given for, which is a particular order. However order-level risk evaluation is difficult as companies do not wish to share their datasets with funders. Furthermore, Small-to-Medium Enterprises (SMEs) themselves may not have enough data to conduct order-level risk evaluation. We propose a Federated Learning (FL) framework to overcome these issues, opening up the possibility for order-level risk evaluation. FL allows collective, order-level model training whilst preserving privacy of data owners. A case study in the aerospace industry indicates that FL can be applied to predict buyers’ late payment risk with minimal performance loss.