Broiler weight forecasting using dynamic neural network models with input variable selection

Abstract

The global demand for poultry meat is predicted to increase by 18% between 2015–17 and 2027, which motivates the need for better tools for production monitoring, planning and optimization. This paper presents the first results on broiler (chicken for meat production) weight forecasting intended for production planning and monitoring using environmental broiler house conditions – such as heating, ventilation, and temperature. We investigate the dynamic impact of environmental conditions on broiler growth, which is known to be highly significant but unexplored in scientific literature. The forecasting is carried out using ensemble dynamic neural network models trained on past production data with mutual information based input variable selection. To investigate the potential of the proposed method, an extensive case study on almost 3.5 years of industrial farm scale production data from a state-of-the-art broiler house is carried out. The dynamic impact of environmental conditions on broiler growth is found to be significant and useful broiler weight forecasts are obtained – effectively providing a foundation for future research on optimization of broiler production.