![[2602.13045] Geometric Manifold Rectification for Imbalanced Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.13045] Geometric Manifold Rectification for Imbalanced Learning
Summary
The paper presents Geometric Manifold Rectification (GMR), a novel framework addressing imbalanced classification in machine learning by leveraging local geometric priors to enhance decision boundaries and protect minority samples.
Why It Matters
Imbalanced learning is a significant challenge in machine learning, often leading to poor model performance. This research introduces a new approach that not only improves classification accuracy but also safeguards minority class data, making it crucial for applications where data imbalance is prevalent.
Key Takeaways
- GMR utilizes geometric confidence estimation for better decision-making.
- Asymmetric cleaning protects minority samples while addressing majority class issues.
- The framework shows competitive performance against traditional sampling methods.
Computer Science > Machine Learning arXiv:2602.13045 (cs) [Submitted on 13 Feb 2026] Title:Geometric Manifold Rectification for Imbalanced Learning Authors:Xubin Wang, Qing Li, Weijia Jia View a PDF of the paper titled Geometric Manifold Rectification for Imbalanced Learning, by Xubin Wang and 2 other authors View PDF HTML (experimental) Abstract:Imbalanced classification presents a formidable challenge in machine learning, particularly when tabular datasets are plagued by noise and overlapping class boundaries. From a geometric perspective, the core difficulty lies in the topological intrusion of the majority class into the minority manifold, which obscures the true decision boundary. Traditional undersampling techniques, such as Edited Nearest Neighbours (ENN), typically employ symmetric cleaning rules and uniform voting, failing to capture the local manifold structure and often inadvertently removing informative minority samples. In this paper, we propose GMR (Geometric Manifold Rectification), a novel framework designed to robustly handle imbalanced structured data by exploiting local geometric priors. GMR makes two contributions: (1) Geometric confidence estimation that uses inverse-distance weighted kNN voting with an adaptive distance metric to capture local reliability; and (2) asymmetric cleaning that is strict on majority samples while conservatively protecting minority samples via a safe-guarding cap on minority removal. Extensive experiments on multiple benchma...