Ensemble selection based on Discriminant functions in binary classification task

Abstract

The paper describes the dynamic ensemble selection. The proposed algorithm uses values of the discriminant functions and it is dedicated to the binary classification task. The proposed algorithm of the ensemble selection uses decision profiles and the normalization of the discrimination functions is carried out. Additionally, the difference of the discriminant functions is used as one condition of selection. The reported results based on the ten data sets from the UCI repository show that the proposed dynamic ensemble selection is a promising method for the development of multiple classifiers systems.