The motivation for this research was the result obtained earlier by the authors in the field of developing industry models for predicting bankruptcy with high prognostic ability. The article examines the prediction reliability of the financial position of companies in the case of introducing an additional category of financial position that reflects the position between financial solvency and insolvency (bankruptcy). The authors hypothesize that the reliability of models decreases if the requirements for their accuracy increase due to the introduction of an additional category of financial position. Hypothesis testing is performed using a non-entropic approach. This approach should reduce the measure of uncertainty in terms of the uncharacteristic nature of some of the identified features of financial position relative to the initial categories. At the same time, features of financial position are defined as ranges of specific weight of balance sheet items that have positive or negative information importance. Information importance is determined based on the methods of system-cognitive analysis, implemented automatically in the EIDOS X++ system, as well as by reproducing information models using MS Excel tools. Normalization of the informational importance values of features and their interpolation allowed us to obtain functions similar to the membership functions in the theory of fuzzy sets. When constructing membership functions relative to ranges of significant balance sheet items ("Fixed assets", "Inventory", "Accounts Receivable", "Short-Term financial investments", "Retained earnings (uncovered loss)", "Accounts payable"), ranges with zero or insignificant values of characteristic functions corresponding to the initial categories of financial position are identified. This actually meant a high level of uncertainty in the prediction. The authors propose to introduce additional linguistic variables and their corresponding fuzzy sets, whose carriers are the relative scales of the above balance items, this will reduce uncertainty. A total of 5 such fuzzy sets were identified, where the researchers used the concept of "gray zone" as a linguistic variable, which was actually used as a new category of financial position. All calculations are shown on the example of fixed assets. The prognostic ability of models based on an optimized sample, where the category of the position of companies that have at least 3 out of 5 features of the "gray zone" has been replaced, is reduced, as expected, but only slightly. And in the case of reproducing algorithms of system-cognitive analysis using MS Excel tools, there is even an increase in the prognostic ability of one of the models. In fact, the hypothesis that the reliability of models decreases if the requirements for their accuracy increase was not confirmed. From an economic point of view, the theoretical significance of the obtained result is that with the help of a non-entropic approach it was possible to show the need to introduce a new category of financial position. From a mathematical point of view, the theoretical significance lies in the fact that membership functions for linguistic variables are obtained based on real data on the financial position of almost two hundred Russian companies, these reduction functions can be used by specialists in the field of fuzzy set theory in the future. The results obtained are applicable at least for the construction industry, but can also be replicated relative to other sectors of the economy when forming the corresponding samples.
Deterministic Annealing Approach to Fuzzy C-Means Clustering Based on Entropy Maximization