Prediction of breeding values for the milk production trait in Iranian Holstein cows applying artificial neural networks

Abstract

The artificial neural networks, the learning algorithms and mathematical models mimicking the information processing ability of human brain can be used non-linear and complex data. The aim of this study was to predict the breeding values for milk production trait in Iranian Holstein cows applying artificial neural networks. Data on 35167 Iranian Holstein cows recorded between 1998 to 2009 were obtained from the Animal Breeding Center of Iran. Breeding values for the milk production trait were determined using the ASReml univariate animal model with 70% of all data used as training data, 15% as testing data and 15% as validating data, to prevent over-fitting of the artificial neural network. A feed-forward backpropagation multilayer perceptron algorithm with three-layer MLP; including 1 input layer, 1 hidden layer and 1 output layer and four-layer MLP; including 1 input layer, 2 hidden layer and 1 output layer was used. The most influential parameters for input characters in artificial neural network were sire, herd, calving year, twice-daily milking (Milk 2x), calving season and age based month. Breeding values for milk production was used as variable output. For network with 4 layers, the best selected structure for the first lactation trait contained input layers with 6 neurons, first hidden layer with 16 neurons and with 68 epoch, second hidden layer with 6 neurons and with 154 epoch and output layer with 1 neuron. The capability of artificial neural network model was higher and closer to the estimated breeding values; therefore it is possible to apply artificial neural networks, instead of commonly-used procedures for predicting the breeding values for milk production.