Application of artificial intelligence for estimating breeding value of body weight in birth and 3 months age in Kermani sheep breed

The purpose of this study was applying artificial intelligence for estimating breeding values of growth traits in Kermani sheep. The records of 867 lambs including animal sex, maternal age, birth weight and weaning weight (3 months old) were used for this purpose. These data were firstly analyzed using ASReml software and then was pre-processed for applying by the MATLAB software. After initial experiments on appropriate architecture for neural networks, it was found that numbers of neurons for input layer were 3 neurons for birth weight and 6 neurons for 3 months weight. The used model in this study was multilayer perceptron (MLP) and applied algorithm was back-propagation to minimize the mean square error. From total data, 70% used as training, 15% for testing and 15% were considered as validation. During the training process, network learning rate was measured regularly by the objective functions and was used network having the minimum error and maximum accuracy. The MLP neural network with 3 input variables, 8 neurons in hidden layer and with correlation coefficient 0.73 has ability to predict breeding values of birth weight and also the MLP neural network with 6 input variables, 3 neurons in hidden layer and with correlation coefficient 0.74 has ability to predict breeding values of 3 months weight. It can be concluded that artificial neural network has a good ability to predict breeding values of growth traits in Kermani sheep with acceptable speed and accuracy and can be used to estimate breeding values for all traits in livestock.