Mitigating Input Noise in Binary Classification: A Unified Framework with Data Augmentation

Pingbo Hu, Grace Yi

Keywords: classification, noisy input, noisy attribute, noisy feature, mismeasured input, measurement error, supervised learning

Abstract: Classification techniques have achieved significant success across fields such as computer vision, information retrieval, and natural language processing. However, much of this progress assumes input features are error-free -- a condition rarely met in practice. In real-world scenarios, noisy inputs caused by measurement errors are common, leading to biased or suboptimal classification results. This paper presents a unified framework for binary classification with noisy inputs, offering a generalizable solution that applies across various supervised learning algorithms and noise models. We provide a theoretical analysis of the bias introduced by ignoring input noise (also referred to as feature corruption) and identify conditions where this bias can be safely disregarded. To address cases where noise correction is needed, we propose a novel data augmentation-based method to mitigate input noise effects. Our approach is both comprehensive and theoretically grounded, providing practical solutions for improving classification accuracy in noisy data enviroments. Extensive experiments, including analyses of medical image datasets, demonstrate the superior performance of our methods under different noise conditions.

Primary Area: learning theory

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8898