Control Charts for Multivariate Non Normal Data Using Winsorization with Comedian when Outliers are Present in Training Data

Abstract

Robust control charts have been proposed for multivariate data as the traditional control charts are affected by extreme observations. Control charts have also been constructed for multivariate non normal data. Our proposed control chart uses comedian as a outlier detection measure along with winsorization to construct control limits in phase I of the chart so that most of the information that is available is utilized and not discarded because of outlier observations. Simulation and winsorization have been used to generate observations from multivariate skew t and the g and h distributions.