@ Randhir:
It works like this:-
Step 1: Create the collection “Milk” with attributes you are interested in. Remember one for unique sample ID.
Step 2: Scan your samples, adding as much info as you can (maybe just Cow Milk, Goat Milk, and unique ID.
Step 3: Make lab analysis of the attributes per unique sample ID
Step 4: Go back into your sample set and add back the lab values against each sample.
Step 5: Create the models for each attribute, and check that they look OK. Take more samples, and clean bad data as appropriate.
Step 6: Test on fresh samples and compare with lab analysis to confirm that it works.
It is NOT enough to take the nutritional values from the labels, manufacturing variance is too great.
You could use label data as a very rough first try, but be sure to junk that data later.