Evidence and Scenario Sensitivities in Naive Bayesian Classifiers

Empirical evidence shows that naive Bayesian classifiers perform quite well compared to more sophisticated network classifiers, even in view of inaccuracies in their parameters. In this paper, we study the effects of such parameter inaccuracies by investigating the sensi-tivity functions of a naive Bayesian classifier. We demonstrate that, as a consequence of the classifier’s independence properties, these sensitivity functions are highly constrained. We investigate whether the various patterns of sensitivity that follow from these functions support the observed robustness of naive Bayesian classifiers. In addition to the standard sensitivity given the available evidence, we also study the effect of parameter inaccura-cies in view of scenarios of additional evidence. We show that the standard sensitivity functions suffice to describe such scenario sensitivities.