A Comprehensive Survey on Ensemble Methods

Abstract

Imbalance dataset is one of the challenge in machine learning to predict the correct class and one state of art solution is Ensemble method. Ensemble method predicts the correct class by combining the predictions of several models rather than prediction based on a particular model. The objective of this paper is to review the traditional ensemble methods like Bagging, Boosting and Stacking generalization and their use to handle the current challenges of the imbalanced dataset. We also present and compare of different modification of these traditional technique to address the limitations of these methods like low diversification in Bagging and overfitting in Boosting. We discuss some new directions in the field of classifier ensembles based on a range of recently published studies. A variety of prior theoretical works have been reviewed in the paper in order to provide a deeper insight into the ensembles themselves.