Multi-Label Feature Selection Combining Three Types of Conditional Relevance.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• The internet's growth generates massive multi-label data, leading to the curse of dimensionality.
• Feature selection is crucial for dimensionality reduction in such datasets.
• Information theory is a common basis for feature selection methods.

Purpose of the Study:

• To develop a novel feature selection method for multi-label data.
• To address the curse of dimensionality in large datasets.
• To improve the accuracy and efficiency of feature selection.

Main Methods:

• Designed a novel feature relevance term (FR) using three incremental information terms.
• Incorporated a label-related feature redundancy term (LR) to minimize redundancy.
• Proposed the multi-label feature selection method: Feature Selection combining three types of Conditional Relevance (TCRFS).

Main Results:

• The proposed TCRFS method comprehensively considers candidate features, selected features, and label correlations.
• FR term aids in capturing optimal features by examining key aspects.
• LR term effectively reduces unnecessary feature redundancy.

Conclusions:

• TCRFS demonstrates superior performance compared to 6 state-of-the-art multi-label approaches.
• The method was validated on 13 multi-label benchmark datasets across 4 domains.
• TCRFS offers an effective solution for feature selection in massive multi-label data scenarios.