Artificial Neural Networks (ANN), which are biologically inspired tool
s, serve as an alternative to regression analysis for complex data. Ba
sed on CP or proximate analysis (PA) of ingredients, two types of ANN
and linear regression (LR) were evaluated for predicting amino acid le
vels in corn, wheat, soybean meal, meat and bone meal, and fish meal.
The two ANN were a three layer Backpropagation network (BP3), and a Ge
neral Regression Neural Network (GRNN). Methionine, TSAA, Lys, Thr, Ty
r, Trp, and Arg were evaluated and R-2 values calculated for each pred
iction method. Artificial neural network training was completed with N
euroShell 2(R) using Calibration to prevent overtraining. Ninety perce
nt of the data were used as the input for the LR and the two ANN. The
remaining 20% (randomly extracted data) were used to calibrate the per
formance of the ANN. As compared to LR, the R-2 values were largest wh
en PA input and GRNN were used. The BP3 did not consistently improve t
he R-2 values for either CP or PA inputs as compared to LR. Each neura
l net can be incorporated into a computer or spreadsheet program.