Credit Risk Analysis with Machine Learning for Peer-to-Peer Lending

Abstract: In the past decade, the scale and scope of fintech credit have snowballed. The peer-to-peer lending industry can be seen as a complement to the traditional banking system. Improving the performance of lending platforms by increasing the accuracy of credit default predictions can help these platforms establish a decisive advantage in the market. This thesis aims to investigate the application of machine learning techniques to P2P lending default prediction modelling. It will seek to identify the most optimal approach for default prediction using machine learning for a given evaluation metric. This study uses real loan data from LendingClub, a publicly accessible public data source, to conduct its credit analysis. A well-rounded set of evaluation metrics was carefully designed and compared. This study discusses four well-established machine learning techniques: logistic regression, support vector machine, random forest, and K-nearest neighbour algorithm. Logistic regression is considered the most adaptable approach for P2P default estimation among the available evaluation metrics after analysing the modelling results. This thesis is of great relevance in helping P2P platforms to improve their ability to identify the credit risk of borrowers. It can also help improve the success rate of P2P online lending, promote reasonable and effective investment by lenders, and boost the development of the P2P online lending industry.