Comparison between methods of robust estimation for reducing the effect of outliers

Abstract EN

In regression analysis the use of least squares method would not be appropriate in solving problem containing outliers or extreme observations.

So, we need alternative method which .t can overcome this problem, this method of estimation is called robust estimation Where the value of the estimation by using this method is not much affected with outliers In this paper we present comparisons between the method of Least Absolute Deviations (LAD) estimation the method of Least Median of Squares (LMS) estimation, the method of L.easi Quantile of Squares (LQS) estimation, the method of Least Trimmed Squares (ITS) estimation the method of Reweightcd Least Squares (LTS.RLS) estimation, the method of M- Huper (MU) estimation and the method of S-estimation in robust regression to determine a suitable regression model.

The best methods are M-cstimation, which represents an extension of the maximum likelihood method and S-estimation is the development of M-estimation mctnod.

The robust regression methods can considerably improve estimation precision.

These results are confirmed using simulation methods and also applied to actual data.