Direct Orthogonalization: some case studies

The effects of a Direct Orthogonalization before applying PCR and PLS are studied for several data sets. In all cases the number of PLS factors needed to obtain the optimal model decreases but the number of PLS and DO factors together is the same as when PLS alone is used. However, the quality of the calibration model (measured as RMSECV) is usually not better when using DO, nor does the predictive quality (RMSEP) change significantly in most cases. The method may be used, however, to obtain a better understanding of the variation present in the data.