EigenU Europe returns to Copenhagen, DENMARK October 17-20, 2022 Complete Info Here!

Orthogonal Signal Correction

What it does

The function osccalc.m calculates the orthogonal signal correction parameters for preprocessing a spectral data prior to developing a PLS model. In situations where the a PLS model captures a very large amount of predictor block (X) variance in the first factor but gets very little of the predicted variable (y or Y) it can be very helpful. In these cases it is useful to remove extraneous variance from X that is unrelated to y. OSC does this by finding directions in X that describe large amounts of variance while being orthogonal to y. Models are built on OSC corrected data, then the OSC correction is applied before application of the PLS models to new data.

In the example shown below, some NIR of diesel fuels is loaded into the workspace, along with the reference viscosity values. A calibration set is produced, and one and two factor OSC corrections are calculated. PLS models are then built on the raw data, and the OSC corrected data. Note how the PLS model built on OSC corrected data grabs a much larger amount of y variance in the first factor. The correction is then applied to the test data, and new predictions are made from the models. A comparison of the prediction errors on a separate test show that OSC was successful: the prediction errors are lower for the OSC corrected models and data.

Requirements for running osccalc

  • MATLAB 5.0
  • PLS_Toolbox or other PLS code

Developed by:

Barry M. Wise
Eigenvector Research, Inc.
bmw@eigenvector.com

Download osccalc.m

To get it, simply click on osccalc.m for Mac, or osccalc.m for PC. Move the file to a folder on your MATLAB path and you’re done.