A Systematic Study of Online Class Imbalance Learning With Concept Drift

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Online class imbalance learning faces challenges from skewed data distributions.
• Concept drift, the change in data properties over time, exacerbates these issues.
• Few studies address the combined problem of class imbalance and concept drift in data streams.

Purpose of the Study:

• To systematically investigate concept drift in class-imbalanced data streams.
• To provide a comprehensive review of existing research, challenges, and focuses.
• To understand effective strategies for handling concept drift in online learning with imbalanced data.

Main Methods:

• Literature review of current research on class imbalance and concept drift.
• Experimental study to evaluate methods for overcoming concept drift.
• Analysis of data streams with skewed class distributions and evolving concepts.

Main Results:

• Identified key research focuses and open challenges in the field.
• Provided empirical evidence on the effectiveness of different approaches.
• Gained insights into overcoming concept drift in imbalanced online learning.

Conclusions:

• This paper offers the first systematic study on concept drift in class-imbalanced data streams.
• The findings contribute to a better understanding of this complex research area.
• Highlights the need for further research into robust online learning algorithms.