engleski

Nonlinear vs. linear regression models in lactation curve prediction

The research objective was to contrast nonlinear (Wood’s) and linear (Wilmink’s, Ali- Schaeffer’s and Guo- Swalve’s) regression models in terms of lactation curve prediction fit. Total of 304, 569 daily yield records automatically recorded on a 1, 136 first lactation Holstein cows from March 2003 till August 2008 were included in analysis. According to the test date, the age at first calving and to the average milk yield in first 40 lactation days, eight age – measuring season – production subgroups were formed. The parameters of analysed nonlinear and linear models were estimated for each defined subgroup. As models evaluation measures, adjusted coefficient of determination, root mean square error, and average and standard deviation of error, were used. Considering obtained results, in terms of total variance explanation the nonlinear Wood’s model showed superiority above the linear ones. Regarding the evaluation measures based on prediction error amount, any notable difference was not determined between analysed regression models. Application of estimated regression parameters, as well as application of artificial neural network (ANN) in predicting dairy cow’s lactation yield on independent dataset, and comparation to method used by the official milk recording system is necessary before making final decision about selection of model with best predictive capabilities of lactation yield in terms of easily practice application.

dairy cows ; daily yields ; regressions ; lactation curves ; prediction

nije evidentirano