Adaptive graph-evolutionary framework for dynamic feature refinement in multi-label learning

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• High-dimensional multi-label learning faces challenges from feature-label interdependencies, redundancy, and noise.
• Traditional feature selection methods lack adaptability to evolving feature relevance during the learning process.

Purpose of the Study:

• To develop an adaptive graph-evolutionary framework for dynamic feature refinement in multi-label learning.
• To address limitations of static feature selection by modeling feature reduction as a sequential decision process.

Main Methods:

• Constructing a heterogeneous graph to model feature-label-instance interactions.
• Employing a graph neural network for dynamic feature reliability evaluation using confidence and uncertainty.
• Integrating feature signals into an evolutionary optimization mechanism for adaptive feature space refinement.

Main Results:

• Demonstrated consistent improvements over state-of-the-art methods on benchmark datasets.
• Achieved significant reductions in Hamming Loss and Ranking Loss.
• Reported notable gains in Average Precision and F1-score, with robust performance across distributions.

Conclusions:

• The proposed adaptive graph-evolutionary framework effectively refines feature spaces for multi-label learning.
• The dynamic approach offers superior predictive performance and scalability compared to conventional methods.
• The framework provides robust and statistically significant improvements in multi-label classification tasks.