Comparison of the Accuracy of Nonlinear Models and Artificial Neural Network in the Performance Prediction of Ross 308 Broiler Chickens

Abstract

This study aimed to investigate and compare nonlinear growth models (NLMs) with the predicted performance of broilers using an artificial neural network (ANN). Six hundred forty broiler chicks were sexed and randomly reared in 32 separate pens as a factorial experiment with 4 treatments and 4 replicates including 20 birds per pen in a 42-day period. Treatments consisted of 2 metabolic energy levels (3000 and 3100 kcal/kg), 2 crude protein levels (22 and 24%) and two sexes. Ten birds in each pen tagged and their weekly BW records were collected individually to evaluate the accuracy of predicted BW by ANN as an alternative to nonlinear regression models (Logistic, Gompertz, Von Bertalanffy, and Brody). Based on the goodness of fit criteria and error measurement statistics, the NLMs fitted the age-weight data better than ANN. The findings indicated that the performance prediction of broiler chicks using the Gompertz model (R2 = 0.9989) was more accurate than other NLMs (R2 = 0.9628 to 0.9988) and ANN (R2 = 0.95839). Therefore, the application of the Gompertz model is suggested to predict the BW changes of Ross 308 broiler chicks over time.