Statistical modeling of calving ease and stillbirths in dairy cattle using the classification tree technique

Abstract

The objective of this paper was to analyze the effects of non-genetic factors affecting calving ease and stillbirths in the population of 1257 Holstein cows, using classification trees. The trees were developed according to the following splitting criterion: the Pearson chi-squared test, entropy function and the Gini index. The quality of decision tree models were compared, taking into consideration the following criteria: the average squared error, misclassification rate, cumulative lift, Kolmogorov–Smirnov statistics and the area under curve of the Receiver Operating Characteristic. A statistical analysis was conducted using the Enterprise Miner 7.1 software included in the SAS package. The calculated quality criteria of the three models that were developed lead to the conclusion that the classification trees established based on the chi-squared statistic and entropy function most accurately define the variability of calving ease. In turn, in the event of stillbirths, classification tree models were identical regardless of the splitting criterion applied. The ranking of variable importance lead to the conclusion that the body weight of the calves, subsequent lactation, husbandry system, gestation length and the sex of the calf are the most significant factors differentiating calving ease. The research showed that stillbirths were diversified only by the body weight of the calves. The present research demonstrates that a graphic model of classification tree can prove a useful tool in distinguishing the factors responsible for calving. It also makes it possible to clearly indicate the factors affecting calving ease and stillbirths that a farmer may observe.