Detection of difficult conceptions in dairy cows using selected data mining methods

In the present study, the detection of difficult conceptions in dairy cows using selected data mining methods – na ve Bayes classifier (NBC) and regression and classifications trees (CART) is presented. The set of 11 diagnostic variables was used, which included, among others, number of lactation, artificial insemination (AI) season, age of inseminated cow, proportion of HF-genes in cow genotype, sex of calf from preceding calving, length of pregnancy, milk, protein and fat yield. Two conception classes were distinguished: the GOOD class, if a cow conceived after one or two AIs and the “POOR” class, if more than two AIs per conception were required. The models were characterized by capability of predicting the membership of conceptions to either class. Correctness of predictions was 83%. CART proved to be more precise in detecting conceptions of the POOR class (sensitivity) compared with predictions by NBC (P≤0.01). Specificity was similar for both classifiers (90% and 93%). Among the variables determining conception class, calving-to-conception interval, calving interval and the difference between the mean body condition and condition at AI were the most significant variables for CART. Utilization of these classifiers, particularly of CART, may help a breeder to appropriately prepare cows for AI, thus contributing to the improved financial results of a herd.