Agriculture Reference
In-Depth Information
interpretation and modelling capacity in
order to prevent processing subjectivity dur-
ing the training algorithm.
Very little research on modelling animal
growth using ANNs has been carried out.
ANNs offer an alternative to regression ana-
lysis for biological growth modelling. Roush
et al
. (2006) compared broiler growth model-
ling using a Gompertz non-linear regression
equation with neural networks. Model accur-
acy was determined by mean squared error
(MSE), mean absolute deviation (MAD), mean
absolute percentage error (MAPE) and bias.
Relative to training data, the neural-developed
neural network model presented lower MSE,
MAD, MAPE and bias. Also for the valid-
ation data, the lowest MSE and MAD were
observed with the genetic algorithm-developed
neural network. The lowest bias was obtained
with the neural-developed network. As meas-
ured by bias, the Gompertz equation under-
estimated the values, whereas the neural- and
genetic-developed neural networks produced
little or no overestimation of the observed
body weight responses.
Wang
et al
. (2012) compared the rela-
tionship between egg production and the
number of pullets, culled birds and
moulted birds in Taiwan using traditional
regression methods or neural network
models. The results showed that the neural
network model is more accurate than the
traditional regression model for predicting
egg production.
Pandorfi
et al
. (2011) evaluated the pre-
cision of multilayer ANNs with error back-
propagation for the prediction of performance
parameters of pregnant sows based on en-
vironmental and physiological variables.
The authors used a single hidden layer with
sigmoidal tangent activation function. Air
temperature and respiratory frequency were
considered as input variables and weight of
piglet at birth and the number of mummi-
fied piglets as output variables. The trained
network presented excellent generalization
power, allowing the prediction of the re-
sponse parameters. The gestation and farrow-
ing environment characterization was adequate
compared with actual data, presenting few
cases of over- or underestimating values. The
use of this expert system to predict animal
performance is feasible because it showed
good results in this application.
A neural network trained to predict the
presence or absence of ascites in broilers
showed excellent performance (Roush
et al
.,
1996). The network topology consisted of
15
physiological variables as inputs, one hid-
den layer with
16
neurons and an output
layer with two neurons (presence or absence
of ascites). Laboratory results were compared
with neural network responses, and the net-
work was shown to be efficient in detecting
the presence or absence of ascites in broilers
before the occurrence of fluid accumulation.
Neural networks with error back-
propagation were used to predict the per-
formance of developing replacement pullets
belonging to a company in southern Brazil
(Salle
et al
., 2001). The authors concluded
that ANNs are able to explain the perform-
ance parameters of developing layer pullets.
The method aids decision making based on
scientifically determined objective criteria.
In addition, it allows the consequences of
the possible decisions to be simulated, and
shows the proportional contribution of each
variable to the phenomenon evaluated.
Savegnago
et al
. (2011) studied the cap-
acity of neural networks to adapt to datasets
in poultry and animal production areas. Neural
networks were applied to an egg production
dataset and models were fitted to the egg
production curve using two approaches, one
being a linear logistic model, and the other
using two ANN models (multilayer percep-
tron (MLP) and radial basis function). The
MLP neural network had the best fit in the
test and validation phases. The advantage of
using neural networks is that they can be fit-
ted to any kind of dataset and do not require
model assumptions such as those required
in the non-linear methodology. The results
confirm that MLP neural networks may be
used as an alternative tool for describing egg
production. The benefits of the MLP are the
great flexibility and their lack of
a priori
assumptions when estimating a noisy non-
linear model.
Ghazanfari
et al
. (2011) mentioned that
ANNs have been shown to be a powerful tool
for system modelling in a wide range of ap-
plications. They applied a back-propagation