A broad review on class imbalance learning techniques

The imbalanced learning issue is related to the performance of learning algorithms in the presence of asymmetrical class distribution. Due to the complex characteristics of imbalanced datasets, learning from such data need new algorithms and understandings to convert efficient large amounts of initial data into suitable datasets. Although several review papers can be found about imbalanced classification problems, none of them contributed an in-depth review of SVM for imbalanced classification problems. To fill this gap, we present an exhaustive review of existing methods to deal with issues linked with class imbalance learning. The majority of the existing survey addresses only classification tasks. We also describe methods to deal with similar problems in regression tasks. A new taxonomy for class imbalanced learning techniques is proposed and classified into three parts: (1) Data pre-processing, (2) Algorithmic structures, and (3) Hybrid techniques. The advantages and disadvantages of each type of imbalanced learning technique are discussed. Moreover, we explain the main difficulties in distributions of imbalanced datasets and discuss the main approaches that have been proposed to tackle these issues. Finally, to stimulate the next research in this area, we emphasize the main opportunities and challenges, which can be useful in research directions for learning algorithms from imbalanced data.