A Systematic Evaluation of Imbalance Handling Methods in Biomedical Binary Classification
Summary
This paper systematically evaluates five imbalance handling methods (RUS, ROS, SMOTE, re-weighting, direct F1 optimization) on three biomedical datasets (tabular, text, image) using models of varying complexity. Results show that benefits depend on model complexity and data modality, with ROS, re-weighting, and direct F1 optimization being effective for complex models on unstructured data.
View Cached Full Text
Cached at: 05/15/26, 06:27 AM
# A Systematic Evaluation of Imbalance Handling Methods in Biomedical Binary Classification Source: [https://arxiv.org/abs/2605.14147](https://arxiv.org/abs/2605.14147) [View PDF](https://arxiv.org/pdf/2605.14147) > Abstract:Objective: The primary goal of this study was to systematically examine the impact of commonly used imbalance handling methods \(IHMs\) on predictive performance in biomedical binary classification, considering the interplay between model complexity and diverse data modalities\. Material and Methods: We evaluated five representative IHMs: random undersampling \(RUS\), random oversampling \(ROS\), SMOTE, re\-weighting \(RW\), and direct F1\-score optimization \(DMO\), against a raw training \(RAW\) baseline\. The evaluation encompassed three public biomedical datasets: MIMIC\-III \(tabular\), ADE\-Corpus\-V2 \(text\), and MURA \(image\), spanning three common biomedical data modalities\. To assess varying model complexity, we employed a range of architectures, from classical logistic regression and random forest to deep neural networks, including multilayer perceptron \(MLP\), BiLSTM, BERT, DenseNet, and DINOv2\. Results: For simpler models such as logistic regression on tabular data, IHMs yielded no significant advantage over the RAW baseline, aligning with prior findings\. However, clear benefits were observed for more complex models and unstructured data: \(a\) ROS and RW consistently enhanced the performance of powerful models; \(b\) direct F1\-score optimization demonstrated utility primarily for unstructured text and image data; and \(c\) RUS and SMOTE consistently degraded performance and are therefore not recommended\. Conclusion: The effectiveness of IHMs depends on both model complexity and data modality\. Performance gains are most pronounced when leveraging appropriate IHMs, such as ROS, RW, and DMO, on high\-complexity models\. ## Submission history From: Jiandong Chen \[[view email](https://arxiv.org/show-email/77194b39/2605.14147)\] **\[v1\]**Wed, 13 May 2026 21:57:38 UTC \(948 KB\)
Similar Articles
Comparative Evaluation of Machine Learning Approaches for Minority-Class Financial Distress Prediction Under Class Imbalance Constraints
This paper presents a comparative evaluation of classical, ensemble, and neural machine learning approaches for predicting financial distress under severe class imbalance, using SMOTE for oversampling and SHAP for interpretability.
RUBRIC: Realism--Utility Balanced Ranking for Imbalanced Classification
RUBRIC is a generator-agnostic filtering framework for imbalanced classification that selects synthetic samples by balancing realism (via a discriminator) and utility (margin-based scoring), improving F1-macro and recall on benchmarks like credit-card fraud detection.
Class Imbalance and Batch Effects in LLM-Based Screening for Systematic Reviews
This study investigates how class imbalance and batch processing affect LLM-based screening in systematic reviews, revealing that batch processing significantly alters decision-making behavior while prevalence metadata has minimal impact.
Model-Agnostic Meta Learning for Class Imbalance Adaptation
University of Memphis researchers propose HAMR, a model-agnostic meta-learning framework that uses bi-level optimization and neighborhood-aware resampling to adaptively reweight hard examples and minority classes across six imbalanced NLP datasets.
Evaluation design conditions the expert-vs-auto MeSH gap: a controlled comparison of bag-of-words and BiomedBERT on the Cohen benchmark
This paper compares expert-assigned and automatically-assigned MeSH terms as features for systematic review screening classifiers, showing that evaluation design significantly affects measured performance gaps, with canonical designs showing larger gaps that attenuate under alternative designs.