Байесовская логистическая регрессия в задаче обучения распознаванию образов при смещении решающего правила

We can face with the pattern recognition problems where the influence of hidden context leads to more or less radical changes in the target concept.Thispaper proposes the mathematical and algorithmic framework for the concept drift in the pattern recognition problems.The probabilistic basis described in this paper is based on the Bayesian approach by the logistic regression for the estimation of decision rule parameters.The pattern recognition procedure derived from this approach uses the general principle of the dynamic programming and has linear computational complexity in contrast to polynomial computational complexity in general kind of pattern recognition procedure.