Exploring Active Learning Based on Representativeness and Uncertainty for Biomedical Data Classification.

Area of Science:

• Biomedical Informatics
• Machine Learning
• Data Science

Background:

• Biomedical data, including images and genetic sequences, is rapidly increasing.
• High costs associated with human annotation limit the utility of this data.
• Current active learning methods struggle with real-world datasets and underutilize classifier knowledge.

Purpose of the Study:

• To develop an active learning approach to reduce human annotation burden in biomedical data.
• To propose a novel active learning strategy that actively incorporates classifier knowledge.
• To improve the efficiency and accuracy of machine learning models through optimized data annotation.

Main Methods:

• Proposed a novel active learning strategy emphasizing classifier participation in sample selection.
• Integrated classifier's own predictions, uncertainty, and sample representativeness into the selection criteria.
• Conducted experiments using the proposed strategy with various supervised classifiers on real-world datasets.

Main Results:

• The novel active learning approach significantly outperformed state-of-the-art methods.
• Demonstrated superior performance across multiple supervised classifiers.
• Achieved a better trade-off between annotation effort and classification accuracy.

Conclusions:

• The proposed active learning strategy effectively leverages classifier knowledge for more informative sample selection.
• This approach offers a practical solution for annotating large-scale biomedical datasets.
• The method enhances model accuracy while reducing annotation costs.