Abstract—In this paper a new method for calculating competence of a classifier in the feature space is presented. The idea of method is based on relating the response of the classifier with the decision profile of a test object which is evaluated using K nearest objects from the validation set. The measure of competence reflects this relation and rates the classifier with respect to the similarity of its response to the decision profile of a test object in a continuous manner. Two multiclassifier systems (MCS’s) representing the dynamic classifier selection (DCS) and dynamic ensemble selection (DES) strategies are developed using proposed measure of competence. The performance of proposed MCS’s was compared against seven multiple classification systems using six benchmark datasets taken from the UCI Machine Learning Repository and Ludmila Kuncheva Collection. The experimental investigations clearly show the effectiveness of the combining multiclassifier system in dynamic fashion using proposed measure of competence regardless of the ensemble type used (homogeneous or heterogeneous).

 

Index Terms—multiclassifier system, dynamic ensemble selection, measure of competence

Cite: Marek Kurzynski, "On a New Competence Measure Applied to the Combining Multiclassifier System," International Journal of Signal Processing Systems, Vol. 4, No. 3, pp. 185-191, June 2016. doi: 10.18178/ijsps.4.3.185-191