Abstract: The article is devoted to developing methodological approach of categorizing the input variables of economic and mathematical models of evaluating of creditworthiness of commercial banks’ borrowers. The basis of mathematical tools was chosen neural network of type of multilayer perceptron. The object of study is the process of categorizing explanatory variables of scoring models. The subject of study is a set of methods of categorizing and approaches to evaluate their impact on the efficiency of the model of estimation of probability of borrower’s default under the contract. As a result of experimental studies within the confines of the proposed methodological approach it was decided to optimize the input variables partitioning into categories by maximizing the Gini coefficient as a measure of the adequacy of scoring models. A conclusion that a drop in the information value does not always act as an indicator of the deterioration of the classifier was obtained.

In this article it has been expanded a list of recommendations for binning that may be used to build more accurate models of evaluating the creditworthiness of borrowers of commercial banks.