Credit Risk Model Based on Machine Learning Techniques

The paper is devoted to building scorecards, which are now widely used in all economically developed countries. Scoring is very popular in the banking sector, while issuing credit each bank checks potential borrower on the probability of obtaining a default. This article aims at constructing mathematical model for assessing the borrower's creditworthiness based on machine learning techniques. This work was done with the usage of SAS Enterprise Miner software, which is a popular tool in the construction of scoring models. This paper describes the method of constructing the scorecard based on the logistic regression model. Also the work presents the main approaches of forming and studying the characteristics of the borrower during the model construction. In the end of the paper the methods of assessing the quality of the constructed model are analyzed. Keywords: credit risk assessment, credit scoring, scorecard, logistic regression, classification, machine learning, risk management, information value, weight of evidence, quality models, credit rating, SAS Enterprise Miner.