British Poultry Science Volume 49, Number 3 (May 2008), pp. 315—320 Predicting performance of broiler chickens from dietary nutrients using group method of data handling-type neural networks H. AHMADI, M. MOTTAGHITALAB 1 , N. NARIMAN-ZADEH 2 AND A. GOLIAN Centre of Excellence in the Animal Science Department, Ferdowsi University of Mashhad, Mashhad, 1 Faculty of Agriculture, Department of Animal Science, University of Guilan, Rasht, and 2 Faculty of Engineering, Department of Mechanical Engineering, University of Guilan, Rasht, Iran Abstract 1. Successful artificial neural network (ANN) applications have been found for many areas. One sub-model of ANNs is the group method of data handling-type neural networks (GMDH-type NNs). The use of self-organising networks leads to successful application in a broad range of areas. However, the use of such methods is not common in poultry science. 2. Broiler chicken nutrition is recognised as a biological system consisting of a complex set of interconnected variables. The adequate information on nutrients (variables), such as metabolisable energy (ME) and amino acid requirements, can help to establish specific feeding programmes, defining optimal performance and reducing production costs. 3. This study addressed the question of whether GMDH-type NNs can be used to estimate the performance of broiler chickens (output) based on specified variables—inputs (dietary crude protein (CP), ME, ME/CP, methionine (Met), lysine (Lys), ME/Met and ME/Lys)—for a commercial broiler chicken farm. The recorded data from 10 broiler chicken flocks were obtained, from March 2003 to April 2005, corresponding to 52 data lines. 4. The results suggested that the GMDH-type NNs may provide an effective means of recognising the patterns in data and accurately predicting the performance of broiler chickens based on investigating inputs. In addition the polynomial equations obtained can be used to optimise the performance of broilers. INTRODUCTION Responses of broiler chickens to dietary crude protein (CP) and amino acids (AA) depend on dietary energy content obtained. There are biological reasons for treating energy as a special case (MacLeod, 2000): firstly, dietary energy has a major role in the control of food intake since the intake of individual nutrients is strongly influ- enced by the nutrient:energy ratio; secondly, the biological systems of the chicken’s control mechanisms may, in effect, perceive the sub- strates as contributors to energy supply rather than identifying them as specific chemicals. Hence, when considering the effects of nutrition on performance of broiler chickens, several dietary nutrients may influence the breast meat yield, feed:gain ratio, mortality and number of days required to reach market weight; among them, metabolisable energy (ME), CP and AA, such as methionine (Met) and lysine (Lys), are more important (Gous, 1998). Although various systems are used to describe the energy and essential AA require- ments of broiler chickens, predicting the performance from the dietary energy and AA patterns in practice and useful terms is still difficult. This difficulty is partly due to the nonlinearity of growth responses related to changes in dietary nutrients (Hruby et al., 1996; Leeson et al., 1996; MacLeod, 2000). A more useful method is to model the nutrition system, which in turn requires an explicit mathematical input—output relationship. Correspondence to: Dr Hamed Ahmadi, Centre of Excellence in the Animal Science Department, Ferdowsi University of Mashhad, PO Box 91775-1163, Mashhad, Iran. E-mail: hahmadima@yahoo.com Accepted for publication 6th March 2008. ISSN 0007–1668(print)/ISSN 1466–1799 (online)/08/030315—6 ß 2008 British Poultry Science Ltd DOI: 10.1080/00071660802136908