An approach to the selection of structure for integrated rating mechanism to be identified given complete learning dataset is suggested. Theoretical assertions and derived from them constructive algorithm for full binary tree selection are described. Challenges for the extension of the approach suggested to incomplete data sets are outlined.