Uncertainty-Aware Generative Oversampling Using an Entropy-Guided Conditional Variational Autoencoder

TL;DR

This work tackles class imbalance in high-dimensional clinical genomics by introducing LEO-CVAE, a CVAE-based oversampling framework that uses Local Entropy Score (LES) to identify uncertain, boundary regions and apply this signal through a Local Entropy-Weighted Loss (LEWL) and entropy-guided sampling. By focusing learning and generation on hard-to-learn regions, LEO-CVAE improves minority-class discrimination while maintaining overall performance, as demonstrated on TCGA lung cancer and ADNI datasets with superior AUC-ROC and AUPRC metrics and balanced macro/micro outcomes. The approach advances oversampling by integrating uncertainty quantification into both the training objective and the generative process, offering a principled path toward robust imbalanced learning in complex nonlinear manifolds typical of clinical genomics. Practical impact includes improved classifier performance on critical biomedical tasks and a framework adaptable to broader data modalities and future diffusion-based or KDE-enhanced generative methods.

Abstract

Class imbalance remains a major challenge in machine learning, especially for high-dimensional biomedical data where nonlinear manifold structures dominate. Traditional oversampling methods such as SMOTE rely on local linear interpolation, often producing implausible synthetic samples. Deep generative models like Conditional Variational Autoencoders (CVAEs) better capture nonlinear distributions, but standard variants treat all minority samples equally, neglecting the importance of uncertain, boundary-region examples emphasized by heuristic methods like Borderline-SMOTE and ADASYN. We propose Local Entropy-Guided Oversampling with a CVAE (LEO-CVAE), a generative oversampling framework that explicitly incorporates local uncertainty into both representation learning and data generation. To quantify uncertainty, we compute Shannon entropy over the class distribution in a sample's neighborhood: high entropy indicates greater class overlap, serving as a proxy for uncertainty. LEO-CVAE leverages this signal through two mechanisms: (i) a Local Entropy-Weighted Loss (LEWL) that emphasizes robust learning in uncertain regions, and (ii) an entropy-guided sampling strategy that concentrates generation in these informative, class-overlapping areas. Applied to clinical genomics datasets (ADNI and TCGA lung cancer), LEO-CVAE consistently improves classifier performance, outperforming both traditional oversampling and generative baselines. These results highlight the value of uncertainty-aware generative oversampling for imbalanced learning in domains governed by complex nonlinear structures, such as omics data.

Figures (2)

• Figure 1: Training history of the LEO-CVAE model for a representative fold on the TCGA lung cancer dataset.
• Figure 2: Training history of the LEO-CVAE model for a representative fold on the ADNI dataset.