Anomaly Detection in Peer-to-Peer Lending Transaction Records by Visualizing High Dimensional Heterogeneous Data

Peer-to-peer (P2P) lending is a new trend in the individual credit business. However, compared with stock market data or other famous marketing data, the lake of P2P data accumulation causes difficulty when building analyzing models. Furthermore, the market inspection and management or relevant law can hardly catch up with the pace of the developing progress of P2P lending market, which brings high risks and uncertainties in this domain. In this work, we proposed a visualization algorithm-based anomaly detection method for analyzing customers who have failed lending records in an online P2P lending platform. Our anomaly detection method is benefited from the visualization map, which can bring an extremely intuitive sense of normal data and outliers. Our approach made it possible for us to show a reasonable assumption for detecting the occasion of such outliers.