29 ene. 2018

Analyzing soy meal in transmittance (part 10 and last)

This is the last post about analyzing soy meal in transmittance. The last option we make was to make a special selection from the original database to see if we get some improvements in the performance of the calibration, but with this option appears some slope, so I prefer the first option were the statistics were quite good but with a Bias.

With all the database the statistics for the PLS regression (with 4 terms) are:
RMSEP..........1,12
SEP............0,529  (Prediction error bias corrected)
RSQ............0,898
and the XY plot is:

With the samples selected with the approach described in the part 9, the statistics are:
RMSEP.......1,12
SEP.........0,633  (Prediction error bias corrected)
Sres........0,547 (Prediction error Slope/Intercept corrected)
RSQ.........0,898
and the XY plot is:
As we can see the first option is the best, and gives quite aceptable errors for protein in intact soy meal in transmittance.
 
And what about if we calibrate with Win ISI
 and validate with a model with the 5 terms in Win ISI:

Samples used for Statistics           27
Slope                                  0.871
Intercept                              5.283
Bias                                  -0.703
SEP                                    0.927
SEP(C)                                 0.616 (Bias corrected)
RSQ                                    0.864


as we can see there are some differences, but the statistics are quite similar and again we have a bias, and an acceptable error.
As a first step we can work with the model adjusted until new samples from the new instrument, be added to the database an the model is recalculated. 
 
Hope you all these posts and let me know if you need more details in the comments.

No hay comentarios:

Publicar un comentario