ASSESSING CREDIT RISK USING MACHINE LEARNING METHODS

The article is devoted to solving the problem of minimizing credit risks of the bank issuing loans to legal persons. This study presents a process for the development and implementation of risk prediction model, one that builds upon a foundation of statistics, data mining and machine learning principles. Statistical and data mining techniques and methodologies have been discussed in detail. The key phases that were covered are: the analysis of the bank's internal rating model, calculation of its discriminatory power, mathematical model based on the rating model was built to predict creditworthiness of future borrowers and assessment of model quality coefficients, the algorithm of data cleaning from anomalies and comparison of the model results was applied. The proposed approach to assessing the borrower's creditworthiness improved the quality of the rating model, as well as its ability to distinguish between reliable and unreliable clients, and reduced the number of default and no default clients. It has been defined that selected quality indexes of the classification models: Accuracy, Specificity, Sensitivity, Balanced accuracy, ROC-curve, AUROC and Gini index are higher for the improved model. A feature of the proposed approach is the application of machine learning methods to data analysis and processing in software R programming language. The practical value of the conducted study is the acquired knowledge on the importance of correct selection of data characterizing the rating model, which in the future will serve as one of the tools for prevention of losses from non-fulfillment of credit obligations. There is a need to devise strategies to not only identify customers who exhibit negative behavior (nonpayment, fraud), but also deal with them effectively to minimize further loss and recoup any monies owed, as quickly as possible.