Prediction of weaning weight of grazing beef by machine learning


Aurelio Guevara-Escobar
Mónica Cervantes-Jiménez
Vicente Lemus-Ramírez
Adolfo Kunio-Yabuta-Osorio
José G. García-Muñiz


Alfalfa, artificial intelligence, regression.


Objective: To develop and validate models using the variables available at calving to predict the weaning weight (WW) of grazing beef calves.

Design/Methodology/Approach: The WW was modelled using machine learning (ML) algorithms and ordinary least squares (OLS). The model included three variable availability scenarios and the best fit was identified using the coefficient of determination (r2), the mean squared error, and the bias.

Results: ML algorithms achieved a better fit than OLS in all scenarios. ML had a 0.70, 0.67, and 0.78 r2 when the following modelling variables were available: B) dam age at calving and parity, calf sex and weight, weaning age, and calving date; I) in addition to the previous variables, dams’ weight at calving, type of calving, calf and cow racial purity; and A) in addition to the all the previous variables, type of service, cow and sire tags and sire breed.

Study Limitations/Implications: The ML and OLS models were representative of a specific database. Modelling based on regional or national data should be studied. Using the lowest number of variables in this study, ML in scenario B provided an acceptable fitting for the prediction modelling of the WW of grazing beef calves.

Findings/Conclusions: ML performed better than OLS, without causing an overfitting, based on the suitability of the WW predictions regarding a database that was not used to train the model.

Abstract 259 | EARLY ACCESS 10 Downloads 0