基于MultiBoost-LMT算法的供应商信用评价研究

Research of Supplier Credit Score Based on MultiBoost-LMT

供应商违约问题一直是供应链管理模式中的一大难题,建立有效的模型实现较准确的供应商违约预测来协助企业采取应对措施,对于企业竞争致胜具有重要意义。本研究首先对MultiBoost算法的框架进行改进,用LMT算法代替C4.5决策树算法,作为MultiBoost的基分类器,提出MultiBoost-LMT算法,其优点是对样本中的奇异点和异常值不敏感,不易出现过拟合现象,具有更高的泛化能力。其次将MultiBoost-LMT算法应用于供应商信用评价问题,在两个公开的供应商信用数据集上的数值试验表明:与其它算法相比,所提出的MultiBoost-LMT算法能够显著地提高供应商信用分类精度,具有较高的实用价值。

Default of supplier has been regarded as one of the toughest difficulties in supply chain management.How to establish an effective model to handle the default of supplier is a significant work.In this paper,a novel method called MultiBoost-LMT algorithm is presented.Due to the fact that the proposed MultiBoost-LMT can effectively avoid overfitting without the loss of the advantages in reducing the bias and the variance of the classified model,the proposed MultiBoost-LMT can increase the model performance significantly.For verification and illustration,two public available supplier credit datasets are used to test and compare the performance of other machine learning algorithm.The experimental results show the proposed MultiBoost-LMT algorithm can yield better performances compared with other machine learning algorithm listed in this study.