Construction and efficiency analysis of neural network models for assessing the financial condition of enterprises

Abstract

© 2019, Institute of Advanced Scientific Research, Inc.. All rights reserved. This article solves the problem of constructing and analyzing the effectiveness of neural network models for assessing the financial condition of enterprises. The urgency of solving this problem on the basis of intelligent modeling technologies, namely, neural networks, is noted. The construction and use of neural network models make it possible to implement an informational approach to modeling, in which data is the starting point. A neural network, learning from the available data, reproduces the patterns hidden in them. A trained neural network can be effectively used to solve the classification problem of assessing the financial condition of enterprises. The accuracy of the classification depends both on the quality of the source data and on the selected architecture of the neural network model. Therefore, the work performed an analysis of the accuracy and comparison of the constructed models based on the hierarchy analysis method. Calculation of weighting factors for the importance of errors of the 1st and 2nd kind for each financial condition of enterprises, as well as classification errors calculated by the method of hierarchy analysis taking into account weighting coefficients allowed us to choose the best neural network model with a minimum value of the average classification error equal to 2.2 %. The selected model is a two-layer perceptron with 90 neurons in the first hidden layer and 20 neurons in the second. The results of evaluating the accuracy of the model showed its effectiveness and the possibility of practical use as part of intelligent decision support systems for assessing the financial condition of enterprises.