NEURAL NETWORK MODEL OF DIAGNOSTICS OF STAGES OF DEVELOPING CORPORATE BANKRUPTCY

Abstract

The article deals with the problem of developing an information and mathematical model to support decisionmaking on the restructuring of corporate debt in the banking technologies of financial management. The purpose of the article is to create a model that allows diagnostic of the stages of developing corporate crisis in difficult conditions of incomplete and noisy data. The model should serve as a tool for improving the objectivity and quality of decisions on the restructuring of corporate debt. The study was conducted on the basis of neural network modelling and system analysis methods, methods of decision-making theory, a solution of inverse problems of interpretation, i.e. extraction of new knowledge from data. We developed an original method of constructing neural network logistic model of bankruptcies (NNLMB) in the difficult conditions of the simulation. New features of the method, increasing the predictive power of the model, are: 1) optimal selection of factors using Bayesian ensemble of auxiliary neural networks, performing compression of factor space; 2) step compression of factors based on the generalized Harrington desirability function; 3) regularization of the main (working) neural network model on Bayesian ensemble of neural networks. NNLMB is tested on real data from corporations of the construction industry. The number of correctly identified objects on the test set was more than 90% on all neural networks of the ensemble. In NNLMB, a sufficiently high prognostic quality of the neural network model is provided by new features of the method and generates an emergent effect, which was proven in computational experiments: the improvement of the quality of the neural network model by the criterion of correctly identified objects Θ is 3.336 times with the compression of factors by 1.35 times. NNLMB can be applied to a wide range of financial management tasks.