DADIR: Density-Aware Data-level Imbalanced Regression Framework

2026-07-19

AI Summary

This work addresses the degradation of regression performance in sparse regions of continuous target variables caused by imbalanced data density distributions. To mitigate this issue, the authors propose a density-aware imbalanced regression framework that dynamically partitions the target space via Density-Aware Adaptive Partitioning (DAAP). Within each partition, a Density-Regularized Conditional Variational Autoencoder (DR-CVAE) generates synthetic samples in the latent space that preserve local structural and density characteristics. A clustering-guided oversampling strategy is further integrated to enhance representation in underrepresented regions. Notably, the framework operates as a plug-in module without requiring modifications to existing regression models, thereby improving their predictive accuracy in minority regions. Experimental results across multiple imbalanced regression benchmarks demonstrate significant gains in overall performance, with particularly pronounced improvements in data-sparse areas.

Abstract

Imbalanced learning addresses predictive modeling problems with underrepresented regions of the data distribution. Although widely studied in classification, imbalanced regression remains challenging because of continuous target variables and heterogeneous density distributions. Existing data-level methods often rely on fixed target partitioning or synthetic sample generation without jointly considering density variations and local feature-space structure. We propose DADIR, a Density-Aware Data-level Imbalanced Regression framework that exploits density information throughout the balancing process. DADIR comprises three components: (1) Density-Aware Adaptive Partitioning (DAAP), which recursively partitions the target space according to density variations; (2) a Density-Regularized Conditional Variational Autoencoder (DR-CVAE), which preserves sparse-region representations while learning latent features; and (3) latent-space data balancing, which combines feature-level clustering with oversampling to generate structurally consistent synthetic samples. Together, these components identify minority regions more effectively, preserve sparse-region information, and generate realistic synthetic data. The resulting balanced dataset can be used directly with existing regression models without modifying their architecture or learning objective. Experiments on diverse imbalanced regression datasets demonstrate consistent improvements in predictive performance, particularly in underrepresented regions, while also improving overall accuracy.

Problem

• imbalanced regression
• density distribution
• data-level imbalance
• continuous target variables
• sparse regions

Innovation

• imbalanced regression
• density-aware partitioning
• conditional variational autoencoder
• latent-space oversampling
• data-level balancing