News Academy of Engineering Sciences A.M. Prokhorov Annotation << Back CLASSIFICATION OF REALIZATIONS OF MEDICAL AND BIOLOGIC SIGNALS WITH USE OF THE METHOD OF NONLINEAR TRANSFORMATIONS G.G. GALUSTOV In article the quasi-heuristic approach giving solving rules quite comprehensible from the engineering point of view and realised without special technical eff orts is considered. Formalisation of a problem of recognition is based on objectively existing generality of properties of objects of the same class, often defi ned by concept of «similarity». The hypothesis used in work about compactness assumes adequacy of concepts of «similarity» of signals of one class and their geometrical «affi nity» shown in their association in one coupled subset in space of signs. Questions of the integrated description of medical and biologic signals on the basis of nonlinear transformation of initial sets for the purpose of shaping optimum from positions of the put forward criterion of a solving rule are considered. The procedure of determination of the operator of the nonlinear transformation is resulted, allowing to construct an optimum dividing surface in space of signs for what it is not required aprioristic knowledge of density functions of signs of classifi ed signals. Can seem, that use only the fi rst and second moments essentially reduces value of the considered solving rule compared, for example, Bayesian rule as the fullest information on signals is concluded in their multidimensional distributions. However in the majority of practical problems of recognition these distributions are a priori unknown and cannot be received with satisfactory accuracy because of limitation of training sample. Therefore for allocation of the useful information from the point of view of recognition it is possible to use only such statistical characteristics which possess suffi cient stability at their estimation on real training sample. Keywords: training sample, fi nal statistics, signals, casual processes, the operator of transformation, probability density, criterion of the optimality, solving rule, probability of an error, reliability of an estimation, nonlinear transformation. Contacts: E-mail: g.galustov@ya.ru Pp. 03-09.