Robust Sparse PCA with R

Numerous methods have been proposed for improving the interpretability of principal component analysis (PCA) results by setting to zero coefficients in the components and thus creating sparse solutions. Few of these methods are capable for identifying atypical observations in the data and providing sparse and robust to outliers solutions at the same time. And finally, from a practical point of view, the availability of software is of utmost importance for the use of these methods in data analysis. We compare three methods for sparse and robust PCA for which R implementations are available at the Comprehensive R Archive Network (CRAN) and illustrate them on real data examples. For selecting the tuning parameter in these methods an alternative criterion, a robust version of the Index of sparseness is proposed and demonstrated to provide very good results.