The following models were used for prediction: artificial neural networks (ANN), including linear networks (LN), multilayer perceptrons with one (MLP1) and two (MLP2) hidden layers and radial basis function (RBF) networks, decision trees, including classification and regression trees (CART) (Breiman et al., 1984) and chi-square automatic interaction detector (CHAID) (Kass, 1980) as well as multivariate adaptive regression splines (MARS) (Friedman, 1991).
The most influential predictor of SP was LW (LN, MLP1, MLP2, CHAID and MARS) or MOL (RBF).
However, the ANN (in the form of MLP1) used for the same purpose had much better predictive performance (prediction determination coefficient equal to 0.64).
The following types of ANN were investigated in the present study: linear network (LN), multilayer perceptron with one hidden layer (MLP1), multilayer perceptron with two hidden layers (MLP2) and radial basis function (RBF) network.
The best MLP1 (RMSE = 0.3665) had a 5-2-1 structure (the number of neurons in the input, hidden and output layers, respectively), whereas the most effective MLP2 (RMSE = 0.3679) had a 5-3-6-1 structure.
We found the highest sensitivity (Se; the percentage of correctly diagnosed dystocic animals) on the training set (0.8013) for MLP1 and it was significantly different from that for the non-neural classifiers (MARS, NBC, GDA, and LR).
The MNLP model can be approximately reformulated as the following mixed 0-1 linear programming model (MLP1) based on linearization of learning curves and quality constraints:
MLP1: min [summation over (w [member of] W)] [summation over (s [member of] S)] [summation over (t [member of] T)] [c.sub.w] [[omega].sub.wst] (17)
The model MLP1 can be further improved in the following two ways.
The main goal of this study was to establish the algorithm with the best predictive capability among classification and regression trees (CART), chi-square automatic interaction detector (CHAID), radial basis function (RBF) networks and multilayer perceptrons with one (MLP1) and two (MLP2) hidden layers in body weight (BW) prediction from selected body measurements in the indigenous Beetal goat of Pakistan.
The r value ranged from 0.82 (MLP1) to 0.86 (RBF and MR).