Development of a model for the prediction of feed intake by dairy cows. 2. Evaluation of prediction accuracy

In a previous paper we have proposed a new concept of a model for the prediction of feed intake by Holstein Friesian dairy cows (Zom et al., 2011). This model predicts feed intake from feed composition and digestibility and the cow's lactation number, stage of lactation and pregnancy. Contrary to many other often used models, this does not include animal performance (milk yield, bodyweight) to predict feed intake. However, BW and MY are highly correlated with DMI. Therefore, the objective of present study was to evaluate the accuracy and robustness of the novel feed intake model and to compare its accuracy and robustness with four other commonly used models for the prediction of feed intake. An evaluation was performed using an independent dataset containing 8974 weekly means of DMI from 348 individual cows observed in 6 feeding experiments including a wide range of diets and management practices was used in this study. Sub-datasets were formed by combining the DMI data by experiment, lactation number, lactation week, and maize silage to grass silage ratios in order to compare the accuracy of the intake models for different feeding practices and groups of cows using mean square prediction error (MSPE) and relative prediction error (RPE) as criteria. The novel model was most accurate as indicated by the MSPEs and RPEs for the whole dataset and the most of the sub-datasets. The results prove that the model of Zom et al. (2011) is able to predict DMI without the use of milk yield or body weight as inputs. It was concluded that novel model was robust and can be applied to various diets and feeding management situations in lactating HF cows