System of boosting voting with multiple learning algorithms

The Ensemble classifier has been an active research topic in the area of machine learning. In a classification task, the ensemble scheme determines a final class label from several individual results which are usually generated by several individual classifiers according to two major principles. The first is to use multiple learning algorithms to form the set of individual outcomes, and the second is to train a set of data fractions, which are generated from a given training dataset, with a weak learning algorithm to generate the set of results. The advantage of the first principle is on classification stability, whereas that of the second principle is on the accuracy improvement in terms of classification accuracy. Thus far, most studies in the literature have been based on either a study of the two principles of ensemble. In this paper we propose a new ensemble scheme to combine the two principles simultaneously. We evaluate the performance on a classification task. Experimental results on several UCI benchmark datasets demonstrate that the proposed framework achieves improved performance in terms of classification accuracy compared to the conventional approaches.