Default risk prediction on a peer-to-peer lending platform

The inherent high default risk in peer-to-peer (P2P) lending necessitates robust credit risk assessment for sustainable online financial operations. This study addresses this need by developing a default prediction model for P2P borrowers using public data from the Renrendai platform in China. With approximately one million loan records, we built up a back-propagation neural network model and achieved over 85% prediction accuracy. The model was refined through two steps: generating Receiver Operating Characteristic and introducing a novel indicator, SPACE, to identify the optimal threshold interval for the final model. This research presents an enhanced credit evaluation model, offering practical implications for P2P lending risk management.