A Review of Unlabeled and Imbalanced Data Challenges in Machine Learning: Strategies and Solutions

ABSTRACT Machine learning models often face significant challenges while dealing with imbalanced and unlabeled datasets. Addressing these issues is resource‐intensive, requiring comprehensive strategies to navigate their individual complexities and compounded effects. This article explores the dual challenges imposed by class imbalance and the absence of labeled data, along with their individual complexities and combined effects on the performance of the model. This study addresses approaches for handling the imbalance problem in datasets, such as data‐level, algorithm‐level, and deep learning methods. The survey also examines hybrid methodologies that integrate these strategies to tackle the compounded issues effectively. Emerging techniques like Bayesian graph‐based learning, uncertainty‐guided semi‐supervised learning, and self‐supervised approaches are also considered for their potential to address the scalability, noise filtering, and generalization challenges associated with imbalanced and unlabeled datasets. It identified persistent gaps, such as the lack of robust evaluation metrics and the underutilization of dynamic feature extraction techniques, suggesting solutions with advanced machine learning approaches. Additionally, the need for adaptive techniques, such as dynamic class weighting and data‐driven filtering mechanisms, is highlighted to address limitations and improve the scalability of machine learning models in real‐world applications. This article is categorized under: Technologies > Machine Learning Technologies > Classification Technologies > Artificial Intelligence