Position Paper:The Significance of Class Imbalance in Online Semi-Supervised Data Stream Learning

Read full text Summarise with VIRAT
Authors: Hadi Al-Kadhimi, Leandro Lei Minku
Cited by: 0
References: 23

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

An important challenge in machine learning is learning from data streams in changing environments, especially when the data is imbalanced and only a few labels are available. A useful approach is semi-supervised online learning, which can take advantage of many unlabelled examples to help the model adapt over time. However, standard methods often have difficulty when imbalance occurs, since the majority class tends to dominate and the minority class is often ignored. In this paper, we show that the challenge of class imbalance is exacerbated by lack of labels in semi-supervised data stream learning, such that tackling class imbalance is an even more significant issue in this context than in supervised data stream learning. However, very little has been done in this area so far. Our study alerts the community of the need for novel semi-supervised strategies that can benefit from unlabelled examples to cope with class imbalance in streaming scenarios, rather than propagating the bias towards the majority class through these examples.

Fields of Study & Subjects

Data Stream Mining Techniques Imbalanced Data Classification Techniques Domain Adaptation and Few-Shot Learning

Publication details

Journal / source
University of Birmingham Research Portal (University of Birmingham)
Publication date
19 October 2026
Language
EN