First Lactation Milk Yield Predicted by the Heifer’s Growth Curve Derivatives

Abstract

Replacement heifers are regularly weighed to assess their health. These data also predict the milk yield in their first lactation (L). The first derivative of the growth curve represents the weight change rate at a given time. It is interesting to use the higher-order derivatives of one biological process, such as growth, to predict the outcome of another process, like lactation. With 78 records of grazing heifers, machine learning was used to predict the L based on variables calculated during the rearing period, from 3 to 21 months of age, every 3 months: body weight (P), first (1D), and second derivative (2D) of an individually modeled Fourier function. Other variables were the age at effective insemination (AI) and the season of the year when the heifer was born (E). The average deviance of the fitted models represented the goodness of fit. The models were trained using 85% of the records, and the fit was evaluated using the remaining data. The deviance was lower for the models including both derivatives in comparison to the models where the derivatives were not included (p = 0.022). The best models predicted the L using data of heifers at six months of age (r2 = 0.62) and the importance of the variables in the model was 35, 28, 21, and 16% for 1D, AI, 2D, and P, respectively. By utilizing this type of model, it would be possible to select and eliminate excess heifers early on, thereby reducing the financial and environmental costs.

Keywords

• grazing
• replacement cattle
• machine learning