Balancing the Imbalanced: A Comprehensive Survey of Techniques for Handling Class Imbalance in Machine Learning

Class imbalance is a pervasive challenge in machine learning, arising when the distribution of class labels in a dataset is skewed. Standard classifiers trained on imbalanced data tend to be biased toward the majority class, producing poor performance on the minority class — which is often the class of greatest practical interest. This paper presents a comprehensive survey of techniques for balancing imbalanced datasets, organized into three principal categories: data-level methods (oversampling, undersampling, and hybrid strategies), algorithm-level methods (cost-sensitive learning, threshold adjustment, and ensemble approaches), and generative approaches (GANs and variational autoencoders). We critically analyze the strengths, limitations, and computational trade-offs of each category, review appropriate evaluation metrics, and present a comparative analysis across benchmark datasets. Practical guidelines for practitioners and future research directions are also discussed.