LGB-BAG在P2P网贷借款者信用风险评估中的应用

Application of LGB-BAG in Credit Risk Assessment of Borrowers from P2P Lending

立足于P2P平台,利用P2P平台个人借款人的信息建立了一套系统的信用风险评估指标体系来甄别可能违约的借款人。基于LightGBM(一种基于决策树的Boosting模型)和Bagging提出一种新的LGB-BAG模型,有效结合了Boosting和Bagging的优势。结果表明,在决策树的个数增大到一定程度的时候,LGB-BAG的F1均值(预测效果)要高于LightGBM和随机森林;并且LGB-BAG的F1方差也要小于其余两种模型。LGB-BAG的F1均值最高可达到0.71175,且LGB-BAG模型能够显著提高信用风险预测效率。

Based on the P2 P platform, this paper uses the information of the individual borrowers of the P2 P platforms to establish a credit risk assessment indicator system to identify borrowers who may default. Therefore, this paper proposes a new LGB-BAG model based on LightGBM(a tree-based Boosting model) and Bagging, which effectively combines the advantages of Boosting and Bagging. The results show that when the number of decision trees increases to a certain extent, the F1 mean of LGB-BAG is higher than that of LightGBM and random forest;and the F1 variance of LGB-BAG is also smaller than that of other two models. The F1 mean of LGB-BAG can reach up to 0.71175, and the LGB-BAG model could improve identification of credit risks of borrowers more effectively.