Active Learning From Imbalanced Data: A Solution of Online Weighted Extreme Learning Machine.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Active learning enhances model quality and reduces training complexity.
• Imbalanced data distributions significantly disrupt active learning performance.
• Existing imbalanced active learning methods often exhibit low efficacy or high computational cost.

Purpose of the Study:

• To propose an efficient active learning solution for imbalanced datasets using the Extreme Learning Machine (ELM) classification model.
• To address the limitations of current imbalanced active learning approaches regarding performance and efficiency.
• To introduce the Active Online-Weighted ELM (AOW-ELM) algorithm.

Main Methods:

• Detailed analysis of active learning disruption factors in imbalanced datasets.
• Hierarchical clustering for initial instance selection to mitigate missed clusters and cold start issues.
• Weighted ELM (WELM) as the base classifier for impartial instance selection, with an online updated mode.
• Development of a flexible early stopping criterion.

Main Results:

• The proposed AOW-ELM algorithm demonstrates superior effectiveness and efficiency.
• Experimental results on 32 imbalanced binary-class datasets validate the algorithm's performance.
• AOW-ELM outperforms several state-of-the-art active learning algorithms designed for imbalanced scenarios.

Conclusions:

• The AOW-ELM algorithm provides an effective and efficient solution for active learning with imbalanced data.
• The study highlights the importance of addressing data imbalance in active learning.
• The developed method offers a robust alternative to existing approaches for imbalanced classification tasks.