Early prediction of milk yield from Holstein-Friesian cow lactation in Querétaro, Mexico

Resumen

Objective: To obtain a machine learning (ML) model to predict the milk yield adjusted to 305 d (MY305) from the same lactation period.

Design/methodology/approach: A database of test days (TD) was used, made up by 11,892 records of daily milk production from cows with more than 150 days in milk (DIM), from 19 barns in Querétaro, Mexico. The milk production was standardized to specific DIMs (5, 10, 20, 30 and 40) and estimations of MY305 were obtained with these, using ML models. The following were incorporated as explicative variables of the herd: month of birth of the cow, month of start of lactation, number of lactation, number of days for three daily milking events, and the two first linear scores of somatic cells.

Results: The best goodness of fit was achieved with ensemble models, obtaining a deviance of 1503584 in the training with 80% of data chosen randomly, while with 20% of the data reserved to evaluate the deviance model it was 1576776. The rate between data observed and predictions of MY305 of the ensemble models had a coefficient of determination of r² = 0.79 and RMSE of 1256. In the best individual model (deviance of 2281420) of ‘deep learning’ type, the most important variables were daily milk production of 30, 10, 5 and 20 DIM (19.9, 16.6, 16.2 and 12.8%, respectively).

Limitations on study/implications: The value of RMSE is high, although the TD databases are generated systematically and usually there are not many barns represented.

Findings/conclusions: For the database examined, milk production in the early phase of lactation together with a set of automatic learning models resulted in an adequate prediction of MY305.