ABSTRACT
The interpretation and labeling of the principal components, regardless of whether they arise from robust or classical PCA, is often difficult. A source of difficulty is the fact that all the loadings are nonzero, since each PC is a linear combination of all the original variables. The common approach to interpretation of principal components is that of ignoring features associated with loadings that are small in absolute value. This practice can be shown to be misleading and sub-optimal (Cadima and Jolliffe, 1995).
