Comparison of different growth models and artificial neural network to fit the growth curve of Lori-Bakhtiari sheep

The aim of this study was to compare different nonlinear and linear regression models and artificial neural network (ANN) for fitting of growth curve in Lori– Bakhtiari sheep breed. Six nonlinear regression models of Negative Exponential, Brody, von Bertalanffy, Gompertz, Logistic and Richards, and two linear regression models of second and third degree polynomial functions along with ANN were used. In total 29517 body weight records of 6320 lambs collected from birth to yearling were analyzed. The data were collected at the Breeding Station of Lori-Bakhtiari sheep in Shahrekord, Iran. The comparison of the models was carried out using coefficient of determination (R2), error mean square (MSE) mean absolute deviation (MAD) and mean absolute percentage error (MAPE) values. All models investigated in the current study fitted the growth data well in Lori– Bakhtiari sheep, according to different goodness of fit criteria. The results indicated that ANN model generated better growth curve fitting of Lori-Bakhtiari sheep than linear and nonlinear growth models and could be used an as alternative method for growth modeling. Regarding the whole models, the artificial neural network was found to be statistically most appropriate model followed by Brody, third degree polynomial, second degree polynomial, Von Bertalanffy, Gompertz, Richards, Logistic and negative exponential growth models, respectively. However, non-linear growth models used to describe the growth will be applicable than the linear models and artificial neural networks. The nonlinear growth models can summarize the growth phenomena in terms of several parameters, with biological interpretation. Among the nonlinear and linear models, the Brody and third degree polynomial functions were better than other models. Negative correlations between the A and k parameters were obtained in all six nonlinear growth models in this study indicated that the sheep with smaller mature weight will be maturing faster. The analysis of variance on the Brody growth curve parameters showed that year of birth and sex significantly influenced (P<0.01) all growth curve parameters. Age of dam had only a significant effect (P<0.05) on k value and did not contribute to differences in A and B values. Type of birth had significant effect (P<0.01) in the B and k values and did not influence on the A value. The results of this study suggest that the most appropriate growth model of Brody can help in the determination of management problems, regulation of feeding programs, and determination of optimum slaughtering age at the LoriBakhtiari sheep breeding station.