Application of credit‐scoring methods in a decision support system of investment for peer‐to‐peer lending

Abstract

Peer-to-peer lending, as a novel lending model, has challenged investors to make effective investment decisions. Issued loans are grouped into default and nondefault. Therefore, different classification methods can be utilized to predict the status of loans in the future. Our study aim is to propose an investment decision model based on the nondefault loans predicted using three different classifiers, including random forest (RF) that is a multitude of decision trees, support vector machine, and naïve Bayes. In fact, we combine each of these classifiers with the portfolio optimization problem to understand which combination leads to the best portfolio concerning the risk and return. In order to find the best combined model, numerical studies are conducted based on real-world data. In addition, the performances of these classifiers are evaluated based on different performance measures using the 10-fold cross-validation method. The results indicate that combining the RF algorithm and portfolio optimization problem leads to the least level of risk. Moreover, it is concluded that RF is the best classifier in terms of performance analysis.