A Comprehensive Survey on Imbalanced Regression: Definitions, Solutions, and Future Directions
Year: 2025
Type: preprint
Abstract: Imbalanced regression occurs when the distribution of a numeric target variable contains rare values, biasing models toward majority regions. Unlike imbalanced classification, it poses unique challenges due to continuous targets and unclear boundaries between frequent and rare values. This survey synthesizes recent advances, covering problem definitions, evaluation metrics, and state-of-the-art methods, distinguishing between tabular and deep settings. An imbalance coefficient for regression, optionally incorporating relevance, is proposed. A curated benchmark dataset mapping is provided. Open challenges and research directions are identified to advance theoretical understanding and practical solutions.
Source: HAL
Authors: Stocksieker, Samuel, Pommeret, Denys
Institution: Aix-Marseille Université
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Reduced inequalities
Open Access status: green
OpenAlex ID: W4415313801