An active learning approach with uncertainty, representativeness, and diversity.

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Big data from the Internet of Things (IoT) presents significant data classification challenges.
• Existing active learning methods often struggle to balance instance uncertainty and representativeness, neglecting diversity.
• Ad hoc approaches for combining query selection criteria limit the effectiveness of active learning.

Purpose of the Study:

• To develop a novel active learning framework that systematically integrates uncertainty, representativeness, and diversity for improved data classification.
• To provide a robust method for measuring and combining these three crucial criteria in instance selection.
• To address the limitations of current active learning algorithms in handling diverse and informative unlabeled instances.

Main Methods:

• A new active learning framework is proposed, considering uncertainty, representativeness, and diversity.
• Instances' uncertainty and representativeness are used to form an informative set.
• Kernel k-means clustering is employed to filter redundant samples from the informative set, yielding the final queried instances.

Main Results:

• The proposed framework systematically measures and combines uncertainty, representativeness, and diversity.
• Experimental results demonstrate superior performance compared to several state-of-the-art active learning approaches.
• The method effectively identifies informative and diverse instances for labeling.

Conclusions:

• The developed active learning framework offers a systematic and effective solution for big data classification challenges in IoT.
• Integrating uncertainty, representativeness, and diversity leads to significant improvements in active learning performance.
• This approach enhances the efficiency and accuracy of data labeling in machine learning applications.