Partial Multi-Label Feature Selection via Entropy-Weighted Multi-Scale Neighborhood Granular Label Distribution...

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Partial multi-label feature selection deals with data where instances have ambiguous label sets.
• Current methods often rely on single-scale assumptions, missing multi-granularity instance-label relationships.

Purpose of the Study:

• To propose a novel framework, PML-FSMNG, for effective partial multi-label feature selection.
• To address limitations of single-scale modeling in handling ambiguous label data.

Main Methods:

• Integrates entropy-weighted multi-scale neighborhood granules with label distribution learning.
• Constructs multi-scale neighborhood systems and uses Shannon entropy for adaptive fusion of label distributions.
• Employs sparse regression with ℓ2,1-norm and entropy-regularized adaptive graph learning.

Main Results:

• The proposed PML-FSMNG method consistently outperforms state-of-the-art approaches on benchmark datasets.
• Demonstrates the effectiveness of multi-scale modeling in feature selection under label ambiguity.
• Highlights the benefits of entropy-guided adaptive learning for preserving geometric structure.

Conclusions:

• The novel framework effectively addresses label ambiguity in partial multi-label feature selection.
• Multi-scale modeling and entropy-guided adaptive learning are crucial for improving feature selection performance.
• PML-FSMNG offers a robust solution for identifying discriminative features in complex, ambiguous datasets.