Using Gompertz model in comparison with artificial neural network to estimate the growth parameters of the broiler chickens fed with sodium bentonite

This experiment was conducted to determine the effects of sodium bentonite on broiler growth parameters, and to compare prediction ability of the Gompertz model and artificial neural network (ANN). A total of 288 one day-old Ross chicken were used in a completely randomized design with 6 treatments (6 diets) and each treatment were divided into 4 replicates. The treatments were diets including 0, 0.75, 1.5, 2.25, 3 and 3.75% sodium bentonite, which was used as top-dress. Gompertz growth function was used to estimate the growth characteristics and to compare the prediction ability of this function with artificial neural network, R square coefficient (R2), mean absolute deviation (MAD), root mean square error (RMSE) and bias were used as indicators. The treatments caused significant changes in the final weight (Wf), the initial weight of chicks (W0), the time for change of growth curve (T) and the weight in the change of growth curve (Wi) (P<0.05), but constant of exponential growth (b) was not affected by the experimental diets. Growth rate at all ages, except the fourth week, was affected by dietary treatment (P<0.05). The R square coefficient for the artificial neural network (0.9992) was higher than Gompertz equation (0.9925. The value of the MAD, RMSE and bias of ANN was less than the Gompertz equation. Overall, this experiment shows that the performance of the ANN for estimation the weight of broilers at 42 days-old was better than Gompertz equation.