ESTIMATION OF SME CREDIT RISKS BY DATA MINING METHODS

Abstract

The article is devoted to solving the scientific and practical problem of modeling credit risks of borrowers of commercial banks – small and medium enterprises (SME). SME lending in Ukraine is characterized by high risk but the need for lending is increasing and essential, which is due to the socio-economic importance of SMEs. That is why there is a need to use methods and models of intelligent data analysis. Using data mining methods, that are perceptron-type neural networks, logistic regressions and decision trees, researched and analyzed in this paper. The database of bank borrowers was used for the research. In particular, 21 financial and economic indicators of enterprise activity were used for modeling. The article carries out a comparative effectiveness analysis of these tools in solving stated problem. During the research, the general population of data was randomly divided into a general and a test sample, and each of them kept the proportion of default units. The experimental calculations demonstrated the greatest suitability for assessing the risks of lending of enterprises the AI methods, namely neural nets of perceptron type. To compare the results of the models, the following values were used: common accuracy, AUC, GINI, specificity, sensitivity. The most significant parameters for the models are also indicated. The study results in recommendations for the application of the built scoring model in banking in order to reduce the level of SME credit operations riskiness.