Classification of Borrowers Using an Immune Model of Artificial Intelligence in Credit Scoring Tasks

for financial organizations, one of the most important tasks is credit scoring - assessing the credit and solvency of potential customers. Classifying clients into risk groups allows banks to make decisions on whether to issue a loan. An approach is proposed that allows you to use any type of features and achieve high accuracy of predictions on various data sets. The study used open data sets most commonly used by developers of credit scoring systems: Credit Risk Analysis, Credit Risk Dataset and German Credit Data. To classify clients by credit risk level, an artificial immune network model and stable algorithms for its training were used. Existing open data sets are limited in size, and the immune model has improved classification accuracy on small training sets. The following indicators of errors in classifying clients by credit risk level were achieved depending on data sets: Credit Risk Dataset: 0.109; Credit Risk Analysis: 0.031; German Credit Data: 0.18. In comparison with the analyzed publications, the size of the training sample was reduced by 2.5-11 times, while the obtained reliability indicators correspond to the world level or exceed the highest previously achieved.