Model for uniqueness assessing degree and for restoration of weakly defined data based on ART-2 neural network modification

The article examines the problem of analyzing and recovering data in small samples with poorly studied relationships, called weakly defined data, by the authors. A method is proposed based on the well-known neural network classification model ART-2, capable of both direct classification and determining the degree of uniqueness of the input vector about the existing sample, taking into account the characteristics of weakly defined data. A modification of the proposed method has also been developed that makes it possible to restore missing attributes in vectors of weakly defined data in the case of the presence of vectors with complete data in the corresponding class. Numerical experiments were carried out for weakly defined data on the content of metals in the blood of children aged 1 to 14 years living in Kazan. Experiments demonstrated the effectiveness of the developed methods.