Comparative Study of the bandwidth in Nonparametric Density Estimated

摘要：

In the density function estimate, the bandwidth compares with the kernel function, the selection of bandwidth is more important. Three optimal bandwidths (bandwidth of MISE, LNO, and the AMSE standard) that depended on detection function (DF), and LSCV standard dont depend on DF. We evaluate the estimate value by Relative Root Mean Square Error (RRMSE) and Relative Bias (RB), studying the good and bad that these methods. When choose to use Semi-Gauss DF and Gauss kernel, in view of the result of the simulation, it seems that the bandwidth of MISE is suitable. For the h of LSCV, Although RB is the least，But, the RRMSE is bigger than MISE, and a lot of line outside data. Because the h of LSCV does not depend on the DF, when the DF did not know, has very big advantage.

关键词： Bandwidth Kernel Function LSCV Line Transect RRMSE

作者: XIONG Guojing

作者单位: School of Economy and Management, Nanchang University, P.R.China, 330031

会议类型: 国际会议

会议名称: 2008年国际应用统计学术研讨会(2008 International Institute of Applied Statistics Studies)

会议地点: 烟台

会议语种:英文

页码: 1-5

在线出版日期: 2008-08-14（万方平台首次上网日期，不代表论文的发表时间）