DBBoost-Enhancing Imbalanced Classification by a Novel Ensemble Based Technique

Abstract

Classification with imbalanced data-sets has become one of the most popular issues in machine learning, since it prevails in various applications. For binary-class problem, the amount of instances from the majority class is significant larger than that from the minority class. Consequently, traditional classifiers achieve a better performance over the majority class, while unsatisfactory predictive accuracy over the minority class. The emergence of ensemble learning provides a possible solution of solving this concern. And there are many recent researches indicate that the combination of Boosting and/or Bagging with pre-processing techniques is an effective way to enhance the classification performance of imbalanced data-sets. Centered on binary-class imbalanced problem, to overcome the drawbacks of state-of-the-art approaches, this paper introduces a novel technique (DBBoost) based on the combination of AdaBoost with an adaptive sampling approach. Through supporting by statistical analysis, experiments show that DBBoost outperforms the state-of-the-art methods based on ensemble.