Q-values, 68 q-values – BUCHI NIRCal User Manual

Page 193

Advertising
background image

Chemometrics

NIRCal 5.5 Manual, Version A

193

3.18.68

Q-Values

Description

Value between 0 and 1. 0 means a bad 1 would be a good calibration. But 1
cannot be reached.
Q-Values depending on the number of PCs.
There is a warning limit for the stored calibration:

0.8 for qualitative

0.6 for quantitative

Use

Select the number of secondary PC according the highest Q-Value.

Method

PCR / PLS / Cluster (CLU) / SIMCA

Matrices ID

64

Tip

In generell: the highest the Q-Value the better the calibration.

Details

The Q-Value is in the calibration protocol, see explanation: LINK:

Calibration

Protocol Qualitative

It is used for selection of the best calibrations for the automatic Calibration
wizard.

Related Topic

Quant Calib. Statistics

,

Cluster per Property

,

Spectra Residuals too big


Explanation of the Q-Value calculation of quantitative calibration:

Value

Weights

Formula for value

General term

Aspects of a
good
calibration

Detail

wNum
NotC

10

Number of C-Set spectra
with Residual too big

Rejection of
known

Rejection
count of known
should be 0

wNum
NotV

1

Number of V-Set spectra
with Residual too big

Rejection of
unknown

Rejection
count of
unknown
should be 0

wQa

2

Abs (SEE-SEP)/
(Abs(SEP)+1.0)

Relative
Consistency

C-Set and V-
Set should
have similar
low standard
error of
prediction

BIAS
corrected

wQb

2

Abs (VsetBIAS)/
Abs(Range)

Weighted
BIAS

The absolute
V-Set BIAS
should be low

wQc

1

1-VsetRegr

Validity

The V-Set
regression
coefficient
should be near
1

wQd

1

Abs( CsetRegr-VsetRegr)

Comparability

C-Set and V-
Set should
have similar
high
regression

wQe

1

SEP / Abs(Range)

Precision

The V-Set
standard error
of prediction
should be low

BIAS
corrected

wQf

1

Abs(RSS)/Abs(Range)

Weighted
Accuracy

The V-Set
Residual Sum
of squares=
RSS should be
low

not BIAS
corrected


Advertising