Abstract: The article is dedicated to solve the main disadvantage, which arise in the process of constructing and applying neural networks and associated primarily with the selection of optimal internal structure, and for the Kohonen maps – with the number of neurons in the hidden layer. It is shown that the process of optimization of neural networks consists in iterative finding of some parameters that provide an extremum of the quality function, which, as a rule, does not have the property of continuity and smoothness. Therefore, a significant drawback of this approach is the inability to ensure the guarantees of the optimality of applied methods and algorithms.

Accordingly, in the article a step-by-step algorithm for constructing Kohonen maps is developed, which intended to solve the problem of classifying dynamic economic systems in accordance with the chosen criterion. For the first time, it was proposed to use the rank coefficient of concordance as an optimality criterion for constructing a neural network-classifier that characterizes the degree of coherence in a set of input variables.

It was experimentally tested step-by-step algorithm for constructing the Kohonen map, which has the optimal topology according to the chosen criterion and splits the original sample into six groups. In accordance with the values of the concordance coefficient, the reaction of key macroeconomic indicators within the obtained clusters is characterized by a high level of similarity