OMAL: A Multi-Label Active Learning Approach from Data Streams.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Digital advancements generate complex, dynamic data streams.
• Real-world data often exhibits complex types like multi-label attributes.
• Online learning scenarios present challenges in adapting to label correlations and data imbalance.

Purpose of the Study:

• To propose a novel online multi-label active learning (OMAL) algorithm.
• To address the challenges of dynamic label correlations and imbalanced data distributions in online scenarios.
• To reduce labeling consumption in dynamic multi-label learning environments.

Main Methods:

• Developed an OMAL algorithm employing uncertainty and diversity as active query strategies.
• Utilized a classifier chain (CC) for multi-label learning, incorporating a label co-occurrence ranking strategy.
• Integrated weight extreme learning machine (WELM) as the base binary-class classifier to handle imbalanced data.

Main Results:

• The proposed OMAL algorithm demonstrated superior performance compared to static multi-label active learning algorithms.
• Evaluated on ten benchmark multi-label datasets transformed into data streams.
• Achieved significant improvements in both Macro-F1 and Micro-F1 metrics.

Conclusions:

• The OMAL algorithm is effective in dynamic data stream environments.
• The method successfully adapts to variances in label correlations and imbalanced data distributions.
• The proposed approach offers a robust solution for online multi-label learning challenges.