A radial basis function neural network (RBFN) approach was used to develop a multi-input, multi-output model for the effect of diets varying in the percentage of ME provided by protein (% MEP), fat (% MEF), and carbohydrate (% MEC) on live weight gain, protein gain, and fat gain in growing chickens. Thirty-three data lines representing response of the White Leghorn male chickens during 23 to 33 d of age to the diets varying in the % MEP, % MEF, and % MEC were obtained from literature and used to train the RBFN model. The prediction values of the RBFN model were compared with those obtained by multiple regression models to assess the fitness of these 2 methods. The fitness of the models was tested using R2, MS error, mean absolute deviation, residual SD, and bias. The developed RBFN model was used to evaluate the relative importance of each input parameter on chicken growth using a sensitivity analysis method. The calculated statistical values corresponding to the RBFN model showed a higher accuracy of prediction than multiple regression models. The sensitivity analysis on the model indicated that dietary % MEP is the most important variable in the growth of chickens, followed by dietary % MEF and % MEC. It was found that the RBFN model is an appropriate tool to recognize the patterns of input-output data or to predict chicken growth in terms of live weight gain, protein gain, and fat gain given the proportion of dietary percentage of ME intake supplied through protein, fat, or carbohydrates.

Keywords