Competitive Particle Swarm Optimization for Multi-Category Text Feature Selection.

Area of Science:

• Computer Science
• Machine Learning
• Artificial Intelligence

Background:

• Multi-label feature selection is crucial for improving text categorization accuracy.
• Existing methods often hybridize evolutionary and filter approaches but neglect operator effectiveness.
• This can lead to suboptimal feature subsets and reduced performance.

Purpose of the Study:

• To propose a novel hybridization approach for multi-label feature selection.
• To address the issue of degenerated feature subsets caused by conventional methods.
• To enhance the accuracy and effectiveness of text categorization algorithms.

Main Methods:

• A new hybridization strategy based on operator competition is introduced.
• The algorithm selectively applies evolutionary and filter operators based on their relative effectiveness.
• This approach modifies feature subsets dynamically for optimal selection.

Main Results:

• Experimental results on 16 diverse text datasets were analyzed.
• The proposed competitive hybridization method demonstrated superior performance.
• It consistently outperformed conventional hybridization techniques.

Conclusions:

• The novel competitive hybridization approach effectively improves multi-label feature selection.
• Selective operator application based on effectiveness leads to better feature subsets.
• This method offers a significant advancement for text categorization accuracy.