Empirical Performance of Nonparametric Kernel Estimators with Data-Driven Bandwidths for Lifetime Distributions

Abstract

During the past several years, many methods for selecting the smoothing parameter have been added in the field of nonparametric kernel regression. The performance of newly introduced selection methods differs quite a little bit, but it is better to consider the recently developed methods for the comparison. Due to the demand for automatic data driven bandwidth selectors for practical statistics, specifically with lifetime distributions, this article provides a review to explain and compare these methods. We conducted an extensive simulation study along with varied estimators and sample sizes to assess the relative performance of the 20 selection methods. From MSEs of simulation, the suitable selection criteria are provided for a suitable bandwidth selection for both classical and plug-in categories. Also, we illustrated these methods with a real example, where we concluded that GCV and Mallow’s Cp perform better with LOWESS estimator.