Improving the accuracy of detecting steroid abuse in cattle by pairwise learning of serum samples

Issues surrounding the misuse of illegal drugs in animals destined for food production have be an enormous challenge to regulatory authorities charged with enforcing their control. A method has been proposed recently which compared the bovine blood biochemistry profiles between control and treated animals, using the support vector machine (SVM) as the classification tool. Whether an animal has been treated is determined by the classification outcome of the SVM on an individual serum sample taken off the animal. However, the acquisition time of the serum sample is essential in the classification performance of the SVM. Thus, the paper proposed to collect and analyze a pair of samples, in order to obtain at least one sample whose acquisition time resulted in an SVM with the highest sensitivity. The power of the strategy in improving sensitivity was theoretically proven to be up to 0.25 and empirically confirmed on a bovine blood biochemistry data. Furthermore, classification rules of the SVM were proposed to be adapted to meet higher levels of demands on sensitivity. Schemes were described which optimized the time apart between the collection of the two samples and the impact of the proposed strategy on specificity was also investigated.