Analysis of Diversity Assurance Methods for Combined Classifiers

Assuring diversity of classifiers in an ensemble plays a crucial role in the multiple classifier system design. The paper presents a comparative study of selected methods which can assure the diversity by manipulating the individual classifier inputs i.e., they train learner using subspaces of a feature set or they try to exploit local competencies of individual classifier for a given subset of feature space. This work is a starting point for developing new methods of diversity assurance embedded in a multiple classifier system design. All methods had been evaluated on the basis of computer experiments which were carried out on benchmark datasets. On the basis of received results conclusions about the usefulness of examined methods for certain types of problems were drawn.