Usefulness of regression models for 100-day milk yield estimation in dairy cows

Abstract

Twenty three models of approximation functions have been analyzed for accuracy of milk yield estimation in dairy cows over a period of 100 days from the date of parturition. The assessment is based on the method of counts and a comparison of estimation error variance. It was found that an increased number of test day yields markedly improves the accuracy of a 100-day yield estimation. Similarity, between the course of the function and the actual lactation curve no significant affect was found for the estimation of a 100-day milk yield based on test yields. The best results of estimation were obtained using the rescaling function. This method, however, requires data from other cows in order to construct a mean lactation curve. Among functions that only use data corresponding to the cow under study, the best results were obtained using linear regression and the difference of two exponential functions. The models were assessed for quality using data on daily milk yields of pure Holstein-Friesian breed of cows and crossbred cows with a Holstein-Friesian admixture.