Vol: 11, Issue: 3 2021

Page: 203-217

Using Different Threshold Values in Wavelet Reduction Method to Estimate the Nonparametric Regression Model With Correlation in Errors

Mohmmed Salh Abdu Alkareem Mahdi, Dr.Saad Kadem Hamza

The wavelet reduction technique is one of the best techniques used in estimating the nonparametric regression function, but it is affected in the event that the errors are related, so (Jonstone) suggested a level-dependent thresholding method to extract the signal from the associated noise. In this paper, a number of types of thresholds will be selected that reduce the risk criterion in estimating the nonparametric regression function and in the presence of a correlation in errors, and these methods are (False Discovery Rate Thresholding), (Bayesshrink Thresholding) and (Universal Thresholding), as simulation experiments were used using Three test and correlation functions of type (AR(1)), sample sizes (64, 128) and different noise ratios. It was found that the best methods were the (False Discovery Rate Thresholding) method, followed by the (Bayesshrink Thresholding) method, while the comprehensive threshold method declined in light of Correlation problem at sample size (128).