Imbalanced Data Solutions in Machine Learning

The field of machine learning is moving towards addressing the challenges of imbalanced data, where the distribution of target variables or classes is skewed. Researchers are proposing innovative solutions to handle imbalanced regression, small sample imbalance problems, and context-aware object detection. A key trend is the development of adaptive oversampling methods, data transformation frameworks, and robust learning techniques that eliminate the need for bias annotations. These approaches aim to improve model performance, reduce spurious correlations, and produce debiased data distributions. Noteworthy papers include Local Distribution-based Adaptive Oversampling, which achieves state-of-the-art results on imbalanced datasets, and Class-Conditional Distribution Balancing, which offers a simple yet effective robust learning method. Additionally, Opposition-Based Data Transformation and Data Augmentation Partial Least Squares Regression models have shown promise in enhancing classification performance and handling uneven categories.

Sources

Local distribution-based adaptive oversampling for imbalanced regression

A Survey on Small Sample Imbalance Problem: Metrics, Feature Analysis, and Solutions

Context Aware Grounded Teacher for Source Free Object Detection

Boosting Classifier Performance with Opposition-Based Data Transformation

DAPLSR: Data Augmentation Partial Least Squares Regression Model via Manifold Optimization

Feature Mixing Approach for Detecting Intraoperative Adverse Events in Laparoscopic Roux-en-Y Gastric Bypass Surgery

Class-Conditional Distribution Balancing for Group Robust Classification

Built with on top of