Dynamic classifier selection based on imprecise probabilities: A case study for the naive bayes classifier

Abstract

Dynamic classifier selection is a classification technique that, for every new instance to be classified, selects and uses the most competent classifier among a set of available ones. In this way, a new classifier is obtained, whose accuracy often outperforms that of the individual classifiers it is based on. We here present a version of this technique where, for a given instance, the competency of a classifier is based on the robustness of its prediction: the extent to which the classifier can be altered without changing its prediction. In order to define and compute this robustness, we adopt methods from the theory of imprecise probabilities. As a proof of concept, we here apply this idea to the simple case of naive Bayes classifiers. Based on our preliminary experiments, we find that the resulting classifier outperforms the individual classifiers it is based on.