Recurrent Algorithms of Structural Classification Analysis for Complex Organized Information

Abstract

For the structural classification analysis of complex organized information, we propose to use recurrent algorithms of stochastic approximation type. We introduce classification quality functionals that depend on non-normalized and zero moments of probability distribution functions for the probability of sample objects appearing in the classes, as well as the type of optimal classification. We propose a new classification algorithm for this type of classification quality criteria and prove a theorem about its convergence that ensures the stationary value of the corresponding functional. We show that the proposed algorithm can be used to solve a wide class of problems in structural classification analysis.