Prediction of weaning weight of grazing beef by machine learning

Abstract

Objective: Train and validate models with variables available at the time of calving to predict the weaning weight (WW) of grazing beef cattle.

Design/methodology/approach: Machine learning (ML) and ordinary least squares (OLS) algorithms were used to model WW of grazing beef calves. There were three scenarios of variable’s availability for modeling, the model of best fit was identified using the determination coefficient (r²), the mean square error and bias.

Results: ML models were better than OLS in all scenarios. The r² was 0.70, 0.67 and 0.78 for the ML with the following modeling variables available: B) dam age and parturition, sex and birth weight, age at weaning and month of birth; I) additionally, dam’s weight at calving, type of calving, calf and dam racial purity; A) additionally, type of service, cow and sire tags.

Limitations on study/implications: The ML and OLS models were representative of this specific database. Further modeling with regional or national databases are needed. Under scenario B, the ML was better in modeling the WW with basic data.

Findings/conclusions: The ML was superior to the OLS without over fitting, since WW predictions were adequate for data not included in model training.