Class Imbalanced Data: Open Issues and Future Research Directions

Since last two decades, imbalanced data is becoming a hot topic to do research or to determine meaningful results. One of the problems of machine learning and data mining areas is class imbalance. Data sets with imbalances have hindered the efficiency of algorithms for data mining and machine learning (in terms of overall accuracy, decision making). In the big data era, the expansion of data mining and machine learning has raised new challenges with the nature of data. In class imbalanced data, majority class lead to problem, i.e., having an imbalance between minority and majority class samples created several problems for researchers. In result, researchers are unable to learn much from systems or they are unable to find or determine prediction or take decision for respective applications like fraud detection, rare diseases identification/ prediction, approval of credit card, software defect prediction, etc. A survey for class imbalance problem is proposed in this paper with discussing several applications (where this problem getting attention). For solving this famous problem or balance this imbalanced data, three methods like Data-level, algorithm-level and hybrid methods are being considered/ used. Among these methods, a hybrid method is receiving much popularity. This paper also discusses several open issues and challenges (which are required to be developed in near future for efficient/ imbalanced learning). Also, in last several (essential) future research directions have been also discussed in this work, which makes this work as important one for research community.