Fish meal NIR prediction

Nhi's picture

Dear all,
I am getting problem with Fish meal, when I scan a FM sample, GH and NH is not so high (4-6) but moisture is very low (<0.5 with t-statistic is above 5) and NIR protein is lower than wet-chemistry results. 
I don't know why because I think moisture is not difficult to be predicted by NIR? Do these FM samples are special?
Thanks all,

jmhhierro's picture

Dear Nhi
Maybe this sample could be an outlier, in a number of cases 3 has been selected as cut off value for GH distanace

rega anggraini's picture

Dear Nhi,
I agree with jmhhierro, maybe your sample is outlier.
Usually value of GH = 3 and NH = 0.6.
How many fish meal do you use to calibration ?
whether your sample has wet chem value below the range of protein used for calibration?
my advice, you can add new sample fish meal with a variety of proteins value so your range protein in calibration is wider.

dwhopkins's picture

Hi Nhi,
Instead of relying on the statistics, does the spectrum of the sample look reasonable?  Or, is it too high in absorbance or wierd in shape?  My guess is, the spectrum is unusual, which is why the predictions don't fall in the normal range, and the stats are out.  Can you show us the spectra of the unusal sample and a normal one?  Sometimes a FM sample might be ground too long, and then the mash does not load into the cell nicely, and the spectrum will be very high and distorted.
Best wishes,