Modified cross-validation in density estimation (Q1193996)

The author proposed a modified cross-validation method for bandwidth selection in nonparametric kernel density estimation. A \(U\)-statistic was introduced as an estimator of the integrated squared error (ISE) of a kernel estimator. This \(U\)-statistic was considered as the modified cross-validation (MCV) function. The author proposed to choose the minimizer of the MCV function as the bandwidth. This bandwidth selection procedure can be carried adaptively. The consistency of such bandwidth was proved and small sample simulation results of comparing with other cross-validation methods for selecting bandwidth selection are given.