Sergey V. Smirnov, Valentina А. Semenova

Samara Federal Research Scientific Center RAS, Institute for the Control of Complex Systems RAS,

The article deals with the problem of taking into account the existence constraints in the widespread task of identifying the properties inherent in an object (or features, or attributes) from an a priori determined set of properties measured in an object in the case of incompleteness and inconsistency of measurement results. The question is raised about the possibility of increasing the productivity of the author's methodological support developed for solving this problem by reducing the description dimension of the existence constraints of properties measured in an object. A natural model of existence constraints is a set of entities (for example, attributes of an object) with two binary coexistence relations defined on it - incompatibility and conditionality; the dimension of such a model is determined by the number of entities and the quantity of the noted existential relations. It is shown that the desired reduction of the existence constraints model of properties measured in an object is feasible based on the identification of equivalence classes in the set of measured properties and the definition of extended binary incompatibility and conditionality relations on the set of discovered classes. Quantitative assessment of the existence constraints reduction effectiveness was made by computer simulation statistical experiments. Of independent interest is the simulation tactics used, due to the multiply connected structure of existence constraints, as well as the multifaceted use of Höfding's inequality to estimate the appropriate number of statistical tests, which establishes exponentially decreasing estimates of the deviations probability of sums of independent random variables from the average of these sums. The results of the study confirmed the significance of dimension reducing of the existence constraints model and revealed the dependence nature of the gain obtained on the parameters that determine the configuration of these constraints.

attributive object identification, existence constraints, equivalence classes, statistical simulation