Graphical tools for visualizing cellwise and casewise outliers

Abstract

Principal component analysis (PCA) and other dimension reduction methods can be affected by cellwise and casewise outliers. Several approaches have been proposed that downweight outlying cells or cases to ensure a more reliable fitting process. The outputs of these robust methods can be used to detect anomalies by means of graphical displays. Our focus is on new visualizations of deviations from a PCA fit that is robust to both cellwise and casewise outliers, and that provides imputed values. The graphics are illustrated on several real datasets, including video data. The visualizations are implemented in a Shiny app.