Principal components transform-partial least squares: a novel method to accelerate cross-validation in PLS regression

This work proposes a new approach for building PLS regression models, Principal Components Transform-PLS (PCT-PLS), which is based on a full eigen decomposition (NIPALS) of the X matrix before proceeding to the PLS regression. This method dramatically accelerates the cross-validation of the calibration models and is at the same time parsimonious in computer memory requirements. This is most noticeable for the huge data sets that are common nowadays. This new approach preserves all the PLS modeling properties, such as robustness and regression vector interpretability, thus facilitating the application of this new procedure to building calibration models. oposed technique will allow the application of PLS modeling to much larger data sets than was previously feasible.