Experimental data analysis by neural nonparametric methods

Nonparametric methods of experimental data analysis by artificial neural networks are considered. A general data set contains a collection of vector pairs of dependent and independent variables (X₁,...,X_n; Y₁,...Y_m)^α, α=1,..,N about which no additional a priori suggestions are made. A methods library includes cluster analysis by Kohonen self-organization maps and discriminant analysis by multilayer networks with backpropagating errors. The discussed methods are applied to the analysis of a database from the field of radiation biology and medicine including medical observations of two territorial population groups. The aim of the research described was to study differences in complex organism state of various groups representatives probably caused by radiation. Based on the Kohonen map an estimation of classification sensibility degree for variations of certain features is made, the optimal informative set of diagnostic parameters is chosen