Predicting first test day milk yield of dairy heifers

Abstract

Profound physiological changes occur in the body of a dairy cow during the transition period. Most of the metabolic disorders and infectious diseases are diagnosed during this time compromising the lactation overall success. Even though it has been proposed methods to evaluate the transition programme in multiparous dairy cows, the same has not been observed for dairy heifers. The present paper aimed at determining which Dairy Herd Improvement (DHI) metrics have the largest impact on first test day milk yield (FT_MILK) of primiparous dairy cows as well as building and comparing predictive models of FT_MILK from animals in this category. We used information from the first test day of 3267 Holstein primiparous dairy cows collected between 2014 and 2017. Data were split into training (n = 2345) and validation (n = 780) data set that were used to estimate model parameters and to evaluate the models, respectively. Variables associated to FT_MILK were identified using regularized regression via elastic net. Three types of models were evaluated: multivariate linear regression (MLR), random forest (RF), and artificial neural network (ANN). The resulting models were evaluated based on six fit-statistics and 10-fold cross validation. In addition, Pearson correlation coefficient and Lin's concordance correlation coefficient were calculated between predicted and observed values, as well as the comparison between observed and predicted median FT_MILK. Statistical significance was declared at an error level α < 0.05. Fifteen DHI variables were identified as the most influential on FT_MILK. The three models were capable of predicting FT_MILK with a mean absolute error lower than 4.00 kg on validation data set and 10-fold cross validation. There was no significant difference between the observed and predicted median FT_MILK. ANN better-predicted FT_MILK compared with MLR and RF. Future work could focus on studying the variation between herds as a possibility to improve the ANN model.