Multiple linear regression: mlr, 5 multiple linear regression: mlr – BUCHI NIRCal User Manual
Page 53

Chemometrics
NIRCal 5.5 Manual, Version A
53
Radii Blow Up limit.
The number of "Total not identified, Cluster OK (%)" can be reduced by reducing the number of
primary PCs or by increasing the Residual Blow Up limit.
Using SIMCA in application
For the identification of an unknown substance the residual should be below the allowed limit and the
unknown spectrum distance should be smaller as the allowed tolerance sphere to the nearest known
calibration spectrum (inside the "inner space").
In the application mode there are 2 answers possible:
Result
Residual
Distance
Identified
OK
OK
Not identified
not OK
not OK
Not identified
not OK
OK
Not identified
OK
not OK
3.1.5 Multiple Linear Regression: MLR
Multiple Linear Regression is an extension of the linear regression to several dimensions.
The analysis is based on a few selected wavelengths and does not require any PCA calculation. In
this procedure, the properties are calculated through intensity values and correlation coefficients, e.g.
it is valid for two selected wavelengths (I
1
and I
2
).
Where:
Prop
property of the „n“th Spectrum
a
intercept
b
1
correlation coefficient of the first wavelength
I
1
intensity at the selected (first) wavelength
Note: I
1
and I
2
must describe independent characteristics. Select at least 3 wavelengths.
Because with MLR only few wavenumber (min. 3) are used and the rest of the measured 1501 (in
case of NIRFlex N-500) are automatically discarded, this simple method is not suggested to use.
The residual cannot be used for outlier detection during the application because of the extreme
wavelength reduction.
This method is only suggested for filter instruments. For Interferometers (full wavelength range) it is
suggested to use PCR or PLS.