The paper outlines neural networks which may be used in practice for bank credit risks evaluation. It demonstrates the neural networks system to classify borrowers, which may be used by bank officers as the system to support decision-making of lending.
The neural network system is a feedforward neural network of the classical topology. To design a neural network, linear activation functions and sigmoidal activation function are used. In this paper the procedure of sample formation of small Russian businesses and choice of risk-dominant performance is shown.
It is demonstrated that designed neural network is characterized by a high predictive power. The obtained results are compared with results of previous studies in the field of credit risks evaluation by using artificial neural networks.