Credit Risk Estimation With Machine Learning

Peer to-peer credit systems have become widespread in recent days and are emerging as alternatives to the traditional banking system. In this system, people can register to the system and present their resources as loans with interest rate and maturity and they can choose their customers themselves. It is important that the selected customer be given the right amount of credit quickly. For this purpose, credit risks should be determined according to customer characteristics and credit types should be given to each customer group separately according to risk ratio. Repayments can also be made at the desired maturities when loans suitable for credit risk are provided. One of the most widely used peer-to-peer credit systems is the so-called Lending Club. Lending Club data is a free set of data that can be used to conduct credit risk analysis. In this study, using the Lending Club data, Microsoft Azure Machine Learning platform was edited with R language, then machine learning algorithms were used in Regression and Binary Classification types, the data were trained and tested and the results were obtained. Boosted Decision Tree Regression algorithm obtained the most successful results among the regression methods, while the Two Class Boosted Decision Tree algorithm yielded the most successful results in binary classification type. Using the best performance algorithms, credit risk applications can be performed according to the characteristics of the customers with web services. In the graphical analysis performed by data visualization applications such as Plotly along with machine learning, the effect levels of various parameters such as credit type, geographical region, interest rate and income groups can be determined by making bad and good credit determination.