Exploring High-Dimensional Data with Robust Principal Components

Abstract

For high-dimensional data of low sample size it is difficult to compute principal components in a robust way. We mention an algorithm which is highly precise and fast to compute. The robust principal components are used to compute distances of the observations in the (sub-)space of the principal components and distances to this (sub-)space. Both distance measures retain valuable information about the multivariate data structure. Plotting the magnitudes of the distance measures helps to reveal important multivariate data information.