Naive Bayesian classifiers for the clinical diagnosis of Classical Swine Fever

Naive Bayesian classifiers have been successfully applied for solving diagnostic problems in the medical domain, but are relatively new to the veterinary field. To demonstrate their potential, we constructed naive Bayesian classifiers for discriminating between Classical Swine Fever (CSF) infected and non-infected herds. To this end, we used data on 490 herds, collected during the 1997/1998 CSF epidemic in the Netherlands. A full naive Bayesian classifier and a selective one were constructed, and their classification accuracies were compared to that of a previously published diagnostic rule. The full classifier had a higher accuracy than the diagnostic rule; the selective classifier proved to be comparable to the rule. In contrast with the diagnostic rule, the two classifiers had the advantage of taking both the presence and the absence of clinical signs into account, which resulted in more discriminative power