| dc.contributor.author |
Kominakis, AP |
en |
| dc.contributor.author |
Abas, Z |
en |
| dc.contributor.author |
Maltaris, I |
en |
| dc.contributor.author |
Rogdakis, E |
en |
| dc.date.accessioned |
2014-06-06T06:44:58Z |
|
| dc.date.available |
2014-06-06T06:44:58Z |
|
| dc.date.issued |
2002 |
en |
| dc.identifier.issn |
01681699 |
en |
| dc.identifier.uri |
http://dx.doi.org/10.1016/S0168-1699(02)00051-0 |
en |
| dc.identifier.uri |
http://62.217.125.90/xmlui/handle/123456789/2184 |
|
| dc.subject |
Artificial neural networks |
en |
| dc.subject |
Dairy sheep |
en |
| dc.subject |
Lactation milk yield |
en |
| dc.subject |
Prediction |
en |
| dc.subject |
Test-day records |
en |
| dc.subject.other |
Dairy products |
en |
| dc.subject.other |
Data acquisition |
en |
| dc.subject.other |
Neural networks |
en |
| dc.subject.other |
Production yield analysis |
en |
| dc.subject.other |
Agriculture |
en |
| dc.subject.other |
artificial neural network |
en |
| dc.subject.other |
milk |
en |
| dc.subject.other |
prediction |
en |
| dc.subject.other |
sheep |
en |
| dc.subject.other |
yield |
en |
| dc.subject.other |
Ovis aries |
en |
| dc.title |
A preliminary study of the application of artificial neural networks to prediction of milk yield in dairy sheep |
en |
| heal.type |
journalArticle |
en |
| heal.identifier.primary |
10.1016/S0168-1699(02)00051-0 |
en |
| heal.publicationDate |
2002 |
en |
| heal.abstract |
The aim of this study was to test the usefulness of artificial neural networks (ANNs) for predicting lactation as well as test-day milk yield(s) in Chios dairy sheep on the basis of a few (2-4) available test-day records at the beginning of a lactation period. The ANN employed was a neural network-like system with some advantages over other ANNs. No selection of learning coefficients, of the number of hidden layers or of the number of neurons in the layers was required. The effect on the network's predictive ability of the number of records used in the training phase, the number of input variables (i.e. test-day records) and data preprocessing was investigated. Input variables were the county, herd, lactation, lambing month, litter size, milk yield recorder, test day and days in milk (after lambing) when the first milk sample was obtained. Various criteria of goodness of prediction of lactation as well as of test-day yields were used, including Pearson and rank correlations between observed and predicted yields; the average difference between observed and predicted yields; the difference between their standard deviations; the standard deviation of differences between observed and predicted yields, and the ratio between it and the observed mean value. The average difference between observed and predicted yields was generally statistically non-significant (P < 0.05) while predicted standard deviations were underestimated. Values of Pearson and rank correlations between observed and predicted lactation yields ranged from 0.87 to 0.97. In prediction of test-day yields, correlation estimates were generally lower than those obtained in lactation yields and declined as the interval between yields increased. Better predictions were obtained as the number of records used for training increased from 500 to 1000, the number of test-day records increased from 2 to 4, and data preprocessing (i.e. encoding of data) was employed. Training the network for low prediction error of a specific parameter did not improve its overall performance. In contrast, network specialization (i.e. using training data for specific parameters prediction) improved the predictive ability of the parameter in question. Results illustrated the potential effectiveness of ANNs in predicting milk yield in dairy sheep and appeared to justify further pursuit of this research. © 2002 Elsevier Science B.V. All rights reserved. |
en |
| heal.journalName |
Computers and Electronics in Agriculture |
en |
| dc.identifier.issue |
1 |
en |
| dc.identifier.volume |
35 |
en |
| dc.identifier.doi |
10.1016/S0168-1699(02)00051-0 |
en |
| dc.identifier.spage |
35 |
en |
| dc.identifier.epage |
48 |
en |