This paper discusses open challenges and future directions in learning from imbalanced data. Despite over two decades of research, imbalanced learning remains a key area of study. Initially focused on binary classification, the topic has expanded to include classification, regression, clustering, data streams, and big data analytics. The paper identifies seven key areas of research: binary and multi-class classification, multi-label and multi-instance learning, semi-supervised and unsupervised learning, regression, data streams, and big data. It highlights challenges such as class imbalance, data preprocessing, algorithm design, and the need for efficient, adaptive, and real-time methods. The paper emphasizes the importance of understanding the structure of minority classes, handling extreme imbalance, adjusting classifier outputs, and improving ensemble learning. It also discusses the need for new methods in multi-class, multi-label, and multi-instance learning, as well as in clustering and regression. The paper concludes that further research is needed to address these challenges and improve the performance of learning systems in imbalanced scenarios.This paper discusses open challenges and future directions in learning from imbalanced data. Despite over two decades of research, imbalanced learning remains a key area of study. Initially focused on binary classification, the topic has expanded to include classification, regression, clustering, data streams, and big data analytics. The paper identifies seven key areas of research: binary and multi-class classification, multi-label and multi-instance learning, semi-supervised and unsupervised learning, regression, data streams, and big data. It highlights challenges such as class imbalance, data preprocessing, algorithm design, and the need for efficient, adaptive, and real-time methods. The paper emphasizes the importance of understanding the structure of minority classes, handling extreme imbalance, adjusting classifier outputs, and improving ensemble learning. It also discusses the need for new methods in multi-class, multi-label, and multi-instance learning, as well as in clustering and regression. The paper concludes that further research is needed to address these challenges and improve the performance of learning systems in imbalanced scenarios.