A high diversity hybrid ensemble of classifiers

Ensemble has been proved a successful approach for enhancing the performance of single classifiers. But there are two key factors influencing the performance of an ensemble directly: accuracy of each single member and diversity between the members. There have been many approaches used in the literature to create the mentioned diversity. In this paper we add a novel approach, in which classifier type variance is utilized along with feature subset diversification to create a high diversity ensemble of different classifiers and an optimization is conducted on the initial population using a multi-objective evolutionary algorithm. The results of experiment over some standard data sets exhibit the outperformance of the suggested approach in comparison to existing ones in specific situations.