On robust partial least squares (PLS) methods

PLS regression methods have been used in applied fields for two decades. Techniques based on iteratively reweighted regression have appeared in the specialized literature with the contaminated data case. We propose a new robust PLS technique based on statistical procedures for covariance matrix robustification. We select the well-known Stahel–Donoho estimator (SDE). We include computational results comparing performance in terms of the standard PLS PRESS reduction if the robust PLS techniques are used. We use simulated and real data and include computational results showing the better robustness and efficiency for the new robust PLS method.