A timely and accurate estimation of body weight in finishing pigs is critical in determining profits by allowing pork producers to make informed marketing decisions on group-housed pigs while reducing labor and feed costs. This study investigated the usefulness of feeding behavior data in predicting the body weight of pigs at the finishing stage. We obtained data on 655 pigs of three breeds (Duroc, Landrace, and Large White) from 75 to 166 days of age. Feeding behavior, feed intake, and body weight information were recorded when a pig visited the Feed Intake Recording Equipment in each pen. Data collected from 75 to 158 days of age were split into six slices of 14 days each and used to calibrate predictive models. LASSO regression and two machine learning algorithms (Random Forest and Long Short-term Memory network) were selected to forecast the body weight of pigs aged from 159 to 166 days using four scenarios: individual-informed predictive scenario, individual- and group-informed predictive scenario, breed-specific individual- and group-informed predictive scenario, and group-informed predictive scenario. We developed four models for each scenario: Model_Age included only age, Model_FB included only feeding behavior variables, Model_Age_FB and Model_Age_FB_FI added feeding behavior and feed intake measures on the basis of Model_Age as predictors. Pearson's correlation, root mean squared error, and binary diagnostic tests were used to assess predictive performance. The greatest correlation was 0.87, and the highest accuracy was 0.89 for the individual-informed prediction, while they were 0.84 and 0.85 for the individual- and group-informed predictions, respectively. The least root mean squared error of both scenarios was about 10 kg. The best prediction performed by Model_FB had a correlation of 0.83, an accuracy of 0.74, and a root mean squared error of 14.3 kg in the individual-informed prediction. The effect of the addition of feeding behavior and feed intake data varied across algorithms and scenarios from a small to moderate improvement in predictive performance. We also found differences in predictive performance associated with the time slices or pigs used in the training set, the algorithm employed, and the breed group considered. Overall, this study's findings connect the dynamics of feeding behavior to body growth and provide a promising picture of the involvement of feeding behavior data in predicting the body weight of group-housed pigs.

Predicting body weight in growing pigs from feeding behavior data using machine learning algorithms / He Yuqing; Tiezzi Francesco; Howard Jeremy; Maltecca Christian.. - In: COMPUTERS AND ELECTRONICS IN AGRICULTURE. - ISSN 0168-1699. - ELETTRONICO. - 184:(2021), pp. 106085-106099. [10.1016/j.compag.2021.106085]

Predicting body weight in growing pigs from feeding behavior data using machine learning algorithms

Tiezzi Francesco;
2021

Abstract

A timely and accurate estimation of body weight in finishing pigs is critical in determining profits by allowing pork producers to make informed marketing decisions on group-housed pigs while reducing labor and feed costs. This study investigated the usefulness of feeding behavior data in predicting the body weight of pigs at the finishing stage. We obtained data on 655 pigs of three breeds (Duroc, Landrace, and Large White) from 75 to 166 days of age. Feeding behavior, feed intake, and body weight information were recorded when a pig visited the Feed Intake Recording Equipment in each pen. Data collected from 75 to 158 days of age were split into six slices of 14 days each and used to calibrate predictive models. LASSO regression and two machine learning algorithms (Random Forest and Long Short-term Memory network) were selected to forecast the body weight of pigs aged from 159 to 166 days using four scenarios: individual-informed predictive scenario, individual- and group-informed predictive scenario, breed-specific individual- and group-informed predictive scenario, and group-informed predictive scenario. We developed four models for each scenario: Model_Age included only age, Model_FB included only feeding behavior variables, Model_Age_FB and Model_Age_FB_FI added feeding behavior and feed intake measures on the basis of Model_Age as predictors. Pearson's correlation, root mean squared error, and binary diagnostic tests were used to assess predictive performance. The greatest correlation was 0.87, and the highest accuracy was 0.89 for the individual-informed prediction, while they were 0.84 and 0.85 for the individual- and group-informed predictions, respectively. The least root mean squared error of both scenarios was about 10 kg. The best prediction performed by Model_FB had a correlation of 0.83, an accuracy of 0.74, and a root mean squared error of 14.3 kg in the individual-informed prediction. The effect of the addition of feeding behavior and feed intake data varied across algorithms and scenarios from a small to moderate improvement in predictive performance. We also found differences in predictive performance associated with the time slices or pigs used in the training set, the algorithm employed, and the breed group considered. Overall, this study's findings connect the dynamics of feeding behavior to body growth and provide a promising picture of the involvement of feeding behavior data in predicting the body weight of group-housed pigs.
2021
184
106085
106099
He Yuqing; Tiezzi Francesco; Howard Jeremy; Maltecca Christian.
File in questo prodotto:
File Dimensione Formato  
he2021cea.pdf

accesso aperto

Tipologia: Pdf editoriale (Version of record)
Licenza: Open Access
Dimensione 7.99 MB
Formato Adobe PDF
7.99 MB Adobe PDF

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/1258580
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 24
  • ???jsp.display-item.citation.isi??? 22
social impact