Detection of Defaulters in P2P Lending Platforms using Unsupervised Learning

Abstract

The lenders and the borrowers favor the P2P lending platforms unlike the traditional lending as P2P lending framework incurs low cost and quick initiation of loans. However the P2P lending platform suffers from a problem that refers to the default borrowers who can't replay the loans and hence generates the financial loss to the investors. In our research we employed four unsupervised learning techniques 1) self-organizing map 2) density based spatial clustering, 3) elliptic envelope and 4) auto-encoders on the Lending club dataset by reducing the features using recursive feature elimination in order to detect the anomalies in form of default borrowers. Our results show that self organizing map is the best performer in detecting the potential defaulters with precision 0.79 and recall 0.816.