Neural Network Models for Assessing the Financial Condition of Enterprises for Supply Chain

Abstract

The paper deals with the task of assessing the financial condition of enterprises. To solve it, we prove the necessity of building a neural network model for supply chain. A set of financial ratios is defined as the input parameters of the model: the current liquidity ratio of the enterprise, the equity ratio, the equity turnover ratio, and the return on equity ratio. The output parameters were the types of the financial condition of enterprises: an unstable state (regression), a normal state (stable) and an absolutely stable state (progression). The volume of input data for building neural network models for assessing the financial condition of enterprises amounted to 210 records. The construction and evaluation of the effectiveness of neural network models are based on the analytical platform Deductor. There have been built 32 modifications of neural network models with different architectures and trained with different samples formed randomly from the source data. To assess the effectiveness of the models built, a technique has been developed, which includes the stages of testing neural networks, evaluating their accuracy and average classification error taking into account weighting factors assigned by an expert. The results of calculations of errors of the first and second type for each financial condition, as well as the average total classification error,  are presented. The best model with a minimum average classification error, which is a single-layer perceptron with 10 hidden neurons, was chosen. The classification accuracy of the model was about 98%. The neural network model is adequate and can be effectively used to solve the problem of assessing the financial condition of enterprises.