Credit Risk Prediction in Peer-to-Peer Lending with Ensemble Learning Framework

Online Peer-to-Peer (P2P) has become a popular way of lending in recent years. Individuals can borrow or lend money directly through an online P2P platform without the help of institutional intermediaries such as banks. It is very important for the platform to predict the credit risk on whether a potential borrower will repay the loan or not, such that the defaulting of borrowers can be avoided as much as possible to keep the platform running healthily. In this paper the machine learning is adopted to design a prediction process, to solve the credit evaluation problem of online P2P lending. A multi-stage ensemble learning model is proposed to evaluate the borrowers' credits, in which the Gradient Boosting Decision Tree (GBDT) algorithm is used for feature mapping and a special Auto-Encoder is applied to extract the best features. Furthermore, the logical regression algorithm is designed to classify the borrowers. Through computational experiments on the open data set and taking AUC (Area Under Curve) as evaluation index, the effectiveness of the proposed model and algorithms is testified.