In the multivariate calibration wizard cross validation can be used
with PLS calibration models.
This validation process allows automatic quality control of the derived
calibration model by keeping some samples out of the calibration and only
use them for prediction. This procedure is repeated during calculation
of the calibration model until all samples have been kept out once. The
residual variance is then calculated from the prediction residuals showing
the overall quality of the model.
In segmented cross validation, the samples are divided into several
groups. During calculation of the calibration model each group or segment
is kept out at a time. There are as many calculations for the calibration
as segments. This procedure will allow to make predictions on all samples.
A final calibration is then calculated including all samples.
The cross validation method can be selected in the following dialog.
Parameters for the selected method must be configured here.
Contents
The dialog contains a drop down box for selection of the cross validation
method applied in calibration process.
With segmented cross validation methods, two additional parameters for
segment adjustment are available:
Number of segments
This number indicates the number of groups. All samples will be evenly
divided and assigned to one of the groups. The maximum number of groups
is equal to the number of samples.
Samples per segment
The user selects the maximum number of samples assigned to a segment
or group. The number of segments is calculated automatically according
to the specified number of samples per segment.
A list box in the lower part of the dialog shows current settings and
assignment of samples per segment.
The following cross validation methods can be configured:
None
The cross validation is deactivated.
Full
The full cross validation considers all samples during calculation of
the calibration model except one. Only one sample at a time is kept out
respectively. There are as many calibration calculations as samples are
included.
Random
All samples are randomly divided into a given number of segments. Either
the number of segments or the
number of samples per segment
can be adjusted in the dialog. Samples are randomly assigned and the resulting
assignment is shown in the list box below.
Systematic 123123123
All samples are assigned systematically in a pre-defined order following
the number of samples per segment.
Example:
Systematic 111222333
All samples are assigned systematically in a pre-defined order following
the number of samples per segment.
