Loan classification using a feed-forward neural network

Objectives. The purpose of the study is to construct and study the use of a feed-forward neural network to solve the problem of loan classification, as well as to conduct a comparative analysis of the neural networkbased approach with the existing approach based on logistic regression.Methods. Based on a feed-forward neural network using historical data on loans issued, the following metrics are calculated: cost function, Accuracy, Precision, Recall, and measure, calculated on Precision and Recall values. Polynomial parameters and the principal component method are used to determine the optimal set of input data for the studied neural network.Results. The impact of data normalization on the final result was analyzed, the influence of the number of units in the hidden layer on the outcome was evaluated using a two-stage method and the Monte Carlo method, the effect of balanced data use was determined, the optimal threshold value for output layer of the neural network under investigation was calculated, the optimal activation function for the hidden layer units was found, the effect of increasing input indicators through missing values imputation and the use of polynomials of varying degrees was studied and the redundancy in the existing set of input indicators was analyzed.Conclusion. Based on the results of the research, we can conclude that the use of a direct distribution network to solve problems of loan classification is appropriate. Compared to logistic regression, implementing a solution using a feed-forward neural network requires more time and computing resources. However, the obtained most important values of Accuracy and measure are higher than those calculated using logistic regression [1].