The relationship between net analyte signal/preprocessing and orthogonal signal correction algorithms

Abstract

A comparison is made between two versions of net analyte signal/processing (NAS/NAP) algorithms and two different orthogonal signal correction (OSC) algorithms. Although the NAP and OSC algorithms preprocess spectra differently, we show by comparison of the external prediction errors (RMSEP) of these algorithms based on terpolymer and synthetic data that these two types of algorithms have the same predictive performance. Under some conditions of overlap and noise, the algorithms can be made to show a slight difference in performance: in those cases, the performance of Lorber's NAS algorithm and Fearn's OSC algorithm track closely and that Olivieri's NAP algorithm tracks with the performance of a modified form of Fearn's OSC. The relationship between these two NAP preprocessing methods is also explored to show that the set of NAS spectra generated by Olivieri's NAP is a subset of that generated by Lorber's, since Lorber's NAP imposes a precisely orthogonal constraint on the target property and Olivieri's NAP relaxes that restriction. These two NAP algorithms can be applied in a piecewise manner for removing local irrelevant information. We also explore the relationship between local versions of these NAP and OSC algorithms. Piecewise implementations of Olivieri's NAP and the modified version of Fearn's OSC are applied to filter a set of spectra. Results from modeling show that these algorithms also have the same performance. Discontinuities occur in spectra preprocessed either by piecewise versions of Fearn's OSC or Lorber's NAP algorithms.