Application of Smoothing Measures in Nonparametric Kernel Classifiers with Usage of Normal Approximation of Probabilities

摘要

The problem of selection of bandwidth with application of kernel estimates of density of distribution was investigated, where instead of one optimal bandwidth for every density estimate we used results of different measures of smoothing for kernel estimates of density. We propose approach, where instead of unprocessed experimental relations normal approximations of the corresponding probabilities is used, and asymptotic properties were investigated for the corresponding conditions of regularity.