INTERACTIVE VARIABLE SELECTION (IVS) FOR PLS .1. THEORY AND ALGORITHMS

A modified PLS algorithm is introduced with the goal of achieving impr oved prediction ability. The method, denoted IVS-PLS, is based on dime nsion-wise selective reweighting of single elements in the PLS weight vector w. Cross-validation, a criterion for the estimation of predicti ve quality, is used for guiding the selection procedure in the modelli ng stage. A threshold that controls the size of the selected values in w is put inside a cross-validation loop. This loop is repeated for ea ch dimension and the results are interpreted graphically. The manipula tion of w leads to rotation of the classical PLS solution. The results of IVS-PLS are different from simply selecting X-variables prior to m odelling. The theory is explained and the algorithm is demonstrated fo r a simulated data set with 200 variables and 40 objects, representing a typical spectral calibration situation with four analytes. Improvem ents of up to 70% in external PRESS over the classical PLS algorithm a re shown to be possible.