Authors
1
Assistant prof., Dept. of Animal Sciences, Ahar Faculty of Agriculture and Natural Resources, University of Tabriz, Ahar, Iran,
2
Associated Professor, Ahar Faculty of Agriculture and Natural Resources, University of Tabriz, Ahar, I. R. Iran.
3
Ahar Faculty of Agriculture and Natural Resources, University of Tabriz, Ahar, Iran
4
Ahar faculty of agriculture, University of Tabriz
Abstract
Background and Objectives: Strategies should be considered to increase the growth and production of sheep meat in Iran. Weight prediction of sheep helps to determine the optimal time for slaughter as well as the appropriate feeding program. Weight prediction can be investigated using mathematical models describing growth.The purpose of this study was to evaluate the performance artificial neural networks in predicting the weight of Moghani sheeps during the growth period of the animal up to one year of age.
Materials and Methods: In this study, the information related to the weight characteristics of 10726 Moghani sheep from birth to one year old, which were collected during the years 1989 to 2016 in the breeding station of Moghani sheep located in Jafarabad Moghan, Ardabil province, was used. To more investigate the growth curve, a multi-layer perceptron artificial neural network accompanied by the backpropagation algorithm was used in this research. Transfer functions such as tangent axon, sigmoid axon, and hyperbolic linear tangent and training algorithms such as momentum, gradient descent, and Levenberg–Marquardt algorithm were used to design the multi-layer perceptron neural network. After fitting nonlinear models and artificial neural network, goodness-of-fit indices including coefficient of determination R2, MSE and MAE were used to select the best model.
Results: The results of this study showed that in the artificial neural network, with three input variables (sex, recording season and age), the hyperbolic axon tangent function and training algorithm of gradient descent was the best performance, with the explanation coefficient, the average square squares, and the average absolute error of 0.919, 602.60 and 3.50, respectively. In the artificial neural network with four input variables (sex, recording season, birth type and age), 1 hidden layer, axon stimulus function, and momentum learning algorithm, had the best performance so that the explanation coefficient, average error squares, and the an absolute error were 0.923, 123/864 and 2864/864, respectively. In the artificial neural network with five input variables (Sex, season of recording, type of birth, age of mother at birth and age of animal), 1 hidden layer, axon hyperbolic linear tangent stimulus function, and Levenberg–Marquardt algorithm, explanation coefficient, the average square squares, and the mean of the absolute error were 0.928, 0 and 2.754, respectively.
Conclusion: The results of this study showed that the artificial neural network model used in this research, with very high accuracy, has the ability to predict the weight of Moghani sheep during the animal's growth period up to one year of age. So that the correlation coefficients in using three, four and five input variables to predict the weight of Moghani sheep were 0.95, 0.96 and 0.96, respectively.
Keywords
Main Subjects