The Study of Different Types of Kernel Density Estimators

Abstract

One of the most important method of estimating and graphing the underlying density is kernel density estimation (KDE). In this paper, we present basic knowledge of KDE, and simulations were carried out which compare three bandwidth selection methods [Normal rule of thumb (NROT), Least squares cross-validation (LSCV), and Biased cross-validation (BCV)]. Four types of kernel (Standard Normal, Biweight, Laplacian, Rational Quadratic and Circular) are chosen to do the simulation. Results shows that overall LSCV performs best.