Ensemble methods for improving classifier performance

Abstract

In this paper, ensemble methods for different base classifiers are proposed. An ensemble technique is a supervised learning algorithm that combines a group of classifiers in order to acquire an overall model with more exact decisions. The classifiers that are support vector machine (SVM), naive Bayes (NB), and back propagation neural network (BPNN) are trained and tested on different gene expression datasets using both random selection method and k-fold cross-validation method. Both binary-class and multi-class datasets are used for evaluation of effectiveness of the ensemble method. Various publicly available gene expression datasets have been used for experiments in order to find the accuracy and effectiveness of the ensemble technique. Performance of the different classification methods and ensemble methods has been compared by using the accuracy values. The results have shown that the accuracy for the gene expression datasets has been increased by using the ensemble methods.