Reliable and relevant modelling of real world data: a personal account of the development of PLS Regression

Why and how the Partial Least Squares Regression (PLSR) was developed, is h ere described from the author's perspective. The paper outlines my frustrat ing experiences in the 70'ies with two conflicting and equally over-ambitio us and oversimplified modelling cultures - in traditional chemistry and in traditional statistics. It describes my mental progress of first learning t o combine them into least squares "unmixing" of known chemical mixtures, an d later extending this into the "unscrambling" of partially unknown structu res as well. The bi-linear regression framework is summarised in terms of t he development from Principal Component Regression into the PLSR. Finally, the versatility of the PLSR is discussed in light of the urgent need for be tter eduacation in scientific data analysis. (C) 2001 Elsevier Science B.V. All rights reserved.