A Study of the Fuzzy Neural Network Method for Assessing Business Financial Viability

Abstract

The presented scientific article conducts a thorough study of methods for assessing the financial solvency of businesses in the conditions of modern economic instability and a steady trend towards an increase in the number of bankruptcy cases of business entities in Ukraine during 2019–2025. The relevance of the work is due to the need to adapt classical mathematical models to the specifics of domestic statistical data, as the direct use of foreign methods without adjusting the coefficients can distort the real assessment of financial stability. The main attention is paid to a comparative analysis of the effectiveness of Fisher's discriminant analysis model and the neuro-fuzzy classifier. The information base of the study consisted of the annual financial statements of enterprises accumulated by the State Statistics Service of Ukraine. To build the models, six key financial indicators were used, reflecting the ratio of cash flows, net profit, revenue, and assets to liabilities and fixed capital. In the course of the work, a specific score function for discriminant analysis was developed and a neuro-fuzzy system was designed, where the base of logical inference rules is formed directly on the basis of the training sample, which minimizes subjective expert influence. Comparison of the results on training, validation, and test data confirmed the advantage of the neuro-fuzzy approach, which demonstrated higher classification accuracy by 1-2% overall. The authors justified the feasibility of using neuro-fuzzy networks as an effective tool for proactive crisis management and simulation modeling of various business development scenarios. The automation of network architecture optimization using the grammatical evolution method to improve the quality of the resulting rule base was identified as a promising direction for further research