comparison of artificial neural networks with other statistical approaches for the prediction of true metabolizable energy of meat and bone meal

• Published in: Poultry science 2010-07, Vol.89 (7), p.1562-1568
• Main Authors: Perai, A.H, Nassiri Moghaddam, H, Asadpour, S, Bahrampour, J, Mansoori, Gh
• Format: Article
• Language: English
• Subjects: accuracy; Animal Feed; Animals; ash content; Biological Products - metabolism; crude protein; energy content; Energy Metabolism; equations; ether extract; feed rations; ingredients; least squares; meat and bone meal; metabolizable energy; Minerals - metabolism; Models, Biological; neural networks; Neural Networks, Computer; nutritive value; Poultry - growth & development; Poultry - metabolism; prediction; regression analysis; simulation models
• ISSN: 0032-5791; 1525-3171
• DOI: 10.3382/ps.2010-00639

There has been a considerable and continuous interest to develop equations for rapid and accurate prediction of the ME of meat and bone meal. In this study, an artificial neural network (ANN), a partial least squares (PLS), and a multiple linear regression (MLR) statistical method were used to predict the TMEn of meat and bone meal based on its CP, ether extract, and ash content. The accuracy of the models was calculated by R² value, MS error, mean absolute percentage error, mean absolute deviation, bias, and Theil's U. The predictive ability of an ANN was compared with a PLS and a MLR model using the same training data sets. The squared regression coefficients of prediction for the MLR, PLS, and ANN models were 0.38, 0.36, and 0.94, respectively. The results revealed that ANN produced more accurate predictions of TMEn as compared with PLS and MLR methods. Based on the results of this study, ANN could be used as a promising approach for rapid prediction of nutritive value of meat and bone meal.