UDC 004.032.26
SEARCH FOR ACCEPTABLE VALUES OF APPLICATION PARAMETERS
FOR LENDING USING A NEURAL NETWORK
N. I. Tsukanova, Ph.D. (Tech.), associate professor of the Department of Computational and Applied
Mathematic, RSREU, Ryazan, Russia;
orcid.org/0000-0001-7337-8037, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
K. G. Shitova, leading development engineer, Sberbank PJSC; Ryazan, Russia;
orcid.org/0000-0002-7809-564X, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The issues of selecting the parameters of client application for lending in order to transfer the borrower
from the class "denied lending" to the class "loan issuance is allowed" are considered. The aim of the work
is to improve the quality of client's creditworthiness assessment procedure, aimed at increasing the number
of borrowers by agreeing with them and/or giving them recommendations on changing loan application parameters.
To achieve this aim, the following tasks were solved in the work. A deep neural network has been
developed in Python, solving the problem of binary classification of clients by their characteristics into two
classes – (admitted, not admitted) to lending. It is shown that the disadvantage of such a procedure is a large
dropout of customers who could get a loan and bring the income to bank, but on other stricter conditions.
Methodology and algorithm of searching loan application parameters values, allowing the client to be admitted
to lending, are proposed.
Key words: client creditworthiness assessment, loan application, deep fully connected neural networks,
Python language, Keras library, error back propagation algorithm, conjugate vector, target (loss) function