Iterative predictor weighting (IPW) PLS: A technique for the elimination of useless predictors in regression problems

A new method for the elimination of useless predictors in multivariate regr ession problems is proposed. The method is based on the cyclic repetition o f PLS regression. In each cycle the predictor importance (product of the ab solute value of the regression coefficient and the standard deviation of th e predictor) is computed, and in the next cycle the predictors are multipli ed by their importance. The algorithm converges after 10-20 cycles. A reduc ed number of relevant predictors is retained in the final model, whose pred ictive ability is acceptable, frequently better than that of the model buil t with all the predictors. Results obtained on many real and simulated data are presented, and compared with those obtained from other techniques. Cop yright (C) 1999 John Wiley & Sons, Ltd.