Multiclass feature selection with metaheuristic optimization algorithms: a review.

Area of Science:

• Machine Learning
• Optimization
• Data Science

Background:

• Feature selection is critical for machine learning, especially for complex multiclass problems.
• Metaheuristic algorithms are increasingly used for optimization tasks, including feature selection.
• Existing research often focuses on binary classification, leaving multiclass feature selection less explored.

Purpose of the Study:

• To systematically survey literature on metaheuristic algorithms for multiclass feature selection.
• To categorize metaheuristic algorithms based on their behavior.
• To identify challenges and research gaps in this domain.

Main Methods:

• Systematic literature review of metaheuristic algorithms for multiclass feature selection (2000-2022).
• Categorization of algorithms into evolutionary-based, swarm-intelligence-based, physics-based, and human-based.
• Analysis of algorithm applications, variations, and associated multiclass classifiers.

Main Results:

• Identified and categorized various metaheuristic algorithms applied to multiclass feature selection.
• Detailed descriptions of algorithms and their performance in selecting optimal feature subsets.
• Examined application areas and popular multiclass classifiers used in conjunction with these algorithms.

Conclusions:

• Metaheuristic algorithms offer effective solutions for multiclass feature selection, enhancing classifier speed and accuracy.
• Further research is needed to address the identified challenges and explore new algorithmic variations.
• This survey provides a comprehensive overview for researchers and practitioners in the field.