Split Sample Sequential Fences based on Bootstrap Cut Off Points for Identifying Outliers and Parameter Estimations

Abstract

Sequential fences is a simple graphical method that is used for detecting outliers. This method is advantageous to the analysts in constructing fences which adjusted for various sample sizes. It is a helpful way to to detect the single and multiple outliers especially in normal or approximately normal data. However, when the distributions of the data are skewed, sequential fences method tends to result in misleading outcome. This paper proposes solution to deal with this problem. The proposed approach with modified algorithm namely Split sample sequential fences with bootstrap (SSFB) is an alternative way to improve sequential fences which can lead to higher accuracy in the detection of outliers and can be applied to a wide range of distributions data. The validity of the new technique has been checked by constructing fences around the true 95% values of different distributions. It was found that the sequential fences constructed by the modified technique not only can detect the outliers in positively skewed distribution but also has smaller bias and smaller root of mean squares error (RMSE) which proves it superiority on the existing techniques.