NONPARAMETRIC MODEL OF STOCHASTIC DEPENDENCE UNDER CONDITIONS OF LARGE ERRORS IN MEASUREMENT OF FUNCTION

The problem of synthesis of a nonparametric model of multidimensional stochastic dependence under conditions of large measurement errors of the values of the function being restored is considered. The essence of the proposed technique consists in combining the process of screening out large measurement errors with a sequential procedure for synthesizing a model of the desired unambiguous dependence. At the first stage of the procedure, a dependence model is synthesized based on the initial sample in the form of nonparametric regression using kernel estimates of probability densities. Deviations of the values of the reconstructed function from its estimate calculated using the primary model are an additional feature when constructing a modification of the nonparametric model of the desired dependence at the second stage of synthesis. Based on the proposed model, the values of the restored dependence in the initial statistical data with large errors are eliminated...