BUCHI NIRCal - Manual (Page 177)

Chemometrics
NIRCal 5.5 Manual, Version A 177
The optimum number of PC is always given by the smallest number of PC where the PRESS function
for the calibration and for the validation set is approximately equal and minimal.
If the error of the prediction diminishes only very slightly by the addition of another PC, it is not worth
while to add that PC. This is because higher PCs with little influence will often result in a poorer
reproducibility or stability of the calibration.
3.18.28 C-Set Property Dependencies
Description Regression coefficient between all original C-Set spectra properties.
Use Shows linear dependencies between different properties.
Method MLR / PCR / PLS
Matrices ID 36
Tip Use a table (grid) to check internal property dependencies
Details Absolute regression coefficient near 1.0 shows that two properties are
linearly dependent. Only the property of C-Set Spectra take effect.
Related Topic Original Property
3.18.29 C-Set Regression Coefficient
Description Calibration Set Regression Coefficient of Original Property and Predicted
Property (also known as correlation coefficient or Pearson's correlation
coefficient)
Use To compare with the V-Set regression Coefficient
Method PCR / PLS
Matrices ID 18
Tip Should be as close to 1 as possible.
Details Visible on the calibration curve and in the calibration protocol.
Formula
Related Topic V-Set Regression Coefficient , Original Property , Predicted Property
The regression coefficient "r" shows how well the predicted values match with the reference values
(original property values) on average.
The correlation is rated as acceptable when r > 0.9 is achieved (the error of the conventional reference
method goes into the NIR-calibration via the reference values).