An Approach for Assimilatiion of Classifier Ensembles on the Basis of Feature Selection and Diversity by Majority Voting and Bagging

A Classifier Ensemble (CE) efficiently improves the generalization ability of the classifier compared to a single classifier. This paper proposes an alternate approach for Integration of classifier ensembles. Initially three classifiers that are highly diverse and showed good classification accuracy when applied to six UCI (University of California, Irvine) datasets are selected. Then Feature Selection (FS) is done and the ensembles are constructed with and without considering FS. Diversities between the classifiers are measured again taking FS into account. Results show that FS has increased the diversity between the classifiers, which in turn increased the accuracy of the ensemble. Finally integration of the constructed ensembles (with FS) is done by using two methods: Bagging and Majority Voting. It is seen that the results of Integration method have shown up to an average of 5% increase in classification accuracy than the individual classifiers and all the ensembles constructed.