A Comparative Analysis of Loan Requests Classification Algorithms in a Peer-to-Peer Lending Platform

Abstract

The Peer-to-Peer (P2P) on-line lending is an emerging lending modality that brings creditors and borrowers closer while enabling a significant reduction of bureaucracy in the lending process. Despite its appealing, the increase in the demand for this lending modality depends on a rigorous settlement of the risk assignment to the potential borrowers. Considering this issue, this article discusses an experimental analysis of classification methods for P2P on-line lending default prediction. The performed experiments were based on the application of the implemented classification algorithms over the data mass formed by borrowers' profiles and loan history records of the P2P Lending Club platform. As the main contribution, the study revealed that it is possible to obtain satisfactory prediction results with a set of attributes smaller than those that were used in studies previously presented in the literature. In addition, it could be verified that, since the algorithms based on decision trees have proved highly effective, the use of these methods is a feasible approach to support the development of lending negotiation tools.