Robust estimation of principal components: a literature review.

Abstract

In this work we do a short literature review on the most relevant methods for robust estimation of Principal Component Analysis (PCA). In particular, we review methods for PCA that are resistant against rowwise outliers, cellwise outliers and against both rowwise and cellwise outliers. It is well known that classical PCA breaks down in the presence of outliers. In practical applications, we suggest to fit a robust method for PCA estimation that is resistant to rowwise and cellwise outliers. We could later compare this result with the classical fit to evaluate the influence of outliers. Robust methodsfor PCA can also be used to detect outliers.