|
||||||||||
|
DOI: 10.17587/it.27.18-24 N. A. Ignat'ev, D. Sc., Professor, e-mail: ignatev@rambler.ru, M. Y. Lolaev, Teacher, musulmon.lolayev.94@mail.ru, National University of Uzbekistan, Tashkent The synthesis of latent features is considered in order to reduce the size of the space for describing objects of disjoint classes K1, K2. A condition is proposed for the sequential formation of feature sets X(h) I X(n), 2 ? h ? n for linear display of object descriptions from X(h) to R2. Latent features are synthesized from data obtained as a result of preprocessing the values of the initial features. Preprocessing is realized through normalization in [0,1] and data transformation that is invariant to the scales of feature measurements. The correspondence structures of objects in X(h) and R2 are investigated according to the results of data preprocessing and grouping. A measure of compliance of two structures of relations of objects in different features spaces is proposed. The measure of compliance is calculated by splitting the objects of classes into disjoint groups. When splitting into groups, the relation of connectedness of class objects by a defined set of hyper balls is used. |