Irene Hoffmann

Robust dimension reduction methods for regression and discriminant analysis are implemented that yield estimates with a partial least squares alike interpretability. Partial robust M regression (PRM) is robust to both vertical outliers and leverage points. Sparse partial robust M regression (SPRM) is a related robust method with sparse coefficient estimate, and therefore with intrinsic variable selection. For binary classification related discriminant methods are PRM-DA and SPRM-DA.

Fully robust versions of the elastic net estimator are introduced for linear and logistic regression, in particular high dimensional data by Kurnaz, Hoffmann and Filzmoser (2017) <DOI:10.1016/j.chemolab.2017.11.017>. The algorithm searches for outlier free subsets on which the classical elastic net estimators can be applied.