Dynamic Classifier Selection

At present, the usual operation mechanism of multiple classifier systems is the combination of classifier outputs. Recently, some researchers have pointed out the potentialities of “dynamic classifier selection’ as an alternative operation mechanism. However, such potentialities have been motivated so far by experimental results and qualitative arguments. This paper is aimed to provide a theoretical framework for dynamic classifier selection and to define the assumptions under which it can be expected to improve the accuracy of the individual classifiers. To this end, dynamic classifier selection is placed in the general framework of statistical decision theory and it is shown that, under some assumptions, the optimal Bayes classifier can be obtained by selecting non-optimal classifiers. Two classifier selection methods that derive from the proposed framework are described. The experimental results obtained in the classification of remote-sensing images and comparisons among different combination methods are reported.