Compare the predictive power of artificial neural network and multiple linear regression to forecast the yearling weight Raeini goats breed

In order to compare the two methods of artificial neural networks (ANN) and multiple linear regression to predict yearling weight used records of 736 Raieni goat. Effects and variables influencing on yearling weight were; animal sex, type of birth, birth year, birth season, herd and characters related to birth weight, three months weight, six months weight and nine months weight. In order to process of data using a neural network model, applied the 3 MLP with a different number of input type. Data modeling were created by neural network software STATISTICA 10. In multiple regression data analyzed using software SAS 9.1.3 Protable with stepwise regression method and suitable model were selected according to R2, RMSE and bias. The results in this research showed a better accuracy of the ANN systems than regression methods (R-square value made by neural networks MLP1 to MLP3 equal to 0.998, 0.998 and 0.997 and RMSE value were estimated 0.96, 0.97 and 1.22, respectively) for predicting yearling weight of these animals.