p

g

p

pp

ed (on the left side) normally have small cow milk concentration

ge values such as 0, 0.125 and 0.25. Moreover, the second cell on

w attracted two spectra profile with two cow milk concentration

ges as 0.75 and 0.625, which were close to each other. The last

he next top row attracted two spectra as well. The cow milk

ation percentages were 0 and 0.125, which were also close to each

erefore the cow milk percentages may be correlated with spectra

ased on this qualitative model.

A self-organising map constructed based on the self-organising map algorithm

rt-term Fourier transform of the milk data. Each figure printed in a cell of this

for a cow milk concentration percentage of a spectrum.

antitative relationship between the cow milk concentration

ges and the spectra profiles represented by the short-term Fourier

m variables was also examined. A neural network regression

as constructed for this relationship analysis. The R package brnn

to construct a neural network regression model. The Jackknife

used for the generalisation test of the regression model.

ds, the predicted cow milk concentration percentages were

d with the measured cow milk concentration percentages. Figure

ws the outcome of this investigation. It can be seen that the