A novel ensemble selection method for cancer diagnosis using microarray datasets

Ensembles of classifiers have recently proved their efficiency in cancer diagnosis based on microarray datasets. The main performance indicators, namely, accuracy and diversity, present the main focus of study when designing an ensemble. One other important performance indicator is classification robustness. In an attempt to improve the performance of an ensemble, the proposed algorithm presents a variation concerning the diversity method used. The proposed algorithm attempts to enhance the robustness of the classification by searching for an ensemble of diverse classifiers. Also, a comparison of the different diversity methods is presented in order to study their impact on the robustness of the classification. The experiments performed show that the diversity method used in the proposed algorithm outperforms the other diversity methods in terms of accuracy and robustness.