Skip to main content
ARS Home » Plains Area » Clay Center, Nebraska » U.S. Meat Animal Research Center » Reproduction Research » Research » Publications at this Location » Publication #341415

Research Project: Genetic and Genomic Approaches to Improve Swine Reproductive Efficiency

Location: Reproduction Research

Title: Feed-forward and generalised regression neural networks in modelling feeding behaviour of pigs in the grow-finish phase

Author
item Cross, Amanda - South Dakota State University
item Rohrer, Gary
item Brown Brandl, Tami
item Cassady, Joseph - South Dakota State University
item Keel, Brittney

Submitted to: Biosystems Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/7/2018
Publication Date: 9/1/2018
Citation: Cross, A.J., Rohrer, G.A., Brown-Brandl, T.M., Cassady, J.P., Keel, B.N. 2018. Feed-forward and generalised regression neural networks in modelling feeding behaviour of pigs in the grow-finish phase. Biosystems Engineering. 173:124-133. https://doi.org/10.1016/j.biosystemseng.2018.02.005.
DOI: https://doi.org/10.1016/j.biosystemseng.2018.02.005

Interpretive Summary: Feeding patterns in group-housed grow-finishing pigs have been investigated for use in management decisions, identifying sick animals, and determining genetic differences within a herd. Feeding behaviour is dependent on several environmental and genetic factors, including but not limited to temperature, humidity, gender, breed, and time of day. Hence, developing accurate models for predicting swine feeding behaviour has proven to be a challenging task. Machine learning is a type of artificial intelligence that focuses on the development of computer programs that can change and adapt when exposed to new data. ARS scientists have used an electronic system to monitor the feeding behaviour of pigs in the grow-finishing phase and utilized machine learning tools to develop a suitable model for predicting swine feeding behaviour based on temperature and time of day. Large deviations between predicted and observed feeding behaviours during an outbreak of pneumonia demonstrated the potential for the model to be used in the automated detection of disease outbreak and other stress events. This work is the first step in developing a computer-based modeling system for swine feeding behaviour. Future work is expected to lead to the development of software tools that will allow swine producers to utilize real-time feeding behaviour data as an early predictor of illness and stress events at the individual animal level.

Technical Abstract: Feeding patterns of pigs have been investigated for use in management decisions and identifying sick animals. Development of models to predict feeding behaviour has been limited due to the large number of potential environmental factors involved and complex relationships between them. Artificial neural networks have been proven to be an effective tool for mapping complicated, nonlinear relationships between inputs and outputs. However, they have not been applied to feeding behaviour prediction. In this study, we compared the use of feed-forward (FFNN) and generalised regression neural networks (GRNN) in forecasting feeding behaviour of pigs in the grow-finish phase, using time of day and temperature humidity index as inputs. Models were calibrated on data from 1923 grow-finish pigs collected from 2008 to 2014, and their predictive ability was tested using data from four additional grow-finish groups collected from 2014 to 2016. Results indicated that FFNN trained with the LevenbergeMarquardt (LM) and scaled conjugate gradient (SCG) algorithms were the most accurate forecasting models. In three of the four validation groups, models trained with LM and SCG algorithms exhibited strong performance, with correlations between predicted and observed feeding behaviours ranging from 0.623 to 0.754. Large deviations between predicted and observed behaviours in the fourth validation group were probably the result of an outbreak of pneumonia, which demonstrates the potential for the model to be used in automated detection of disease outbreak and other stress events. This work is the first step in developing a fully automated system for detecting changes in feeding behaviour.