Differential Evolution With Duplication Analysis for Feature Selection in Classification.

Area of Science:

• Machine Learning
• Data Science
• Computational Intelligence

Background:

• Feature selection is crucial for reducing data dimensionality and improving classification.
• Existing methods often overlook multiple optimal feature subsets that yield comparable performance.
• Identifying diverse optimal solutions is essential for robust model development.

Purpose of the Study:

• To propose a novel niching-based differential evolution (DE) method for discovering multiple optimal feature subsets.
• To address the limitation of existing methods that focus on single optimal solutions.
• To enhance the diversity and classification accuracy of feature selection.

Main Methods:

• A niching-based differential evolution (DE) algorithm incorporating duplication analysis and a subset repairing scheme.
• An improved mutation operator utilizing both niche and global information for generating promising feature subsets.
• A novel selection method prioritizing diversity among feature subsets for population update.

Main Results:

• The proposed method demonstrated superior classification accuracy compared to seven evolutionary and two traditional feature selection algorithms across 18 datasets.
• The algorithm successfully identified multiple distinct feature subsets achieving very similar or identical classification performance.
• Experimental results validate the effectiveness of the DE-based approach in finding diverse optimal solutions.

Conclusions:

• The proposed niching-based DE method effectively addresses the challenge of multiple optimal feature subsets in feature selection.
• This approach enhances classification performance and provides a set of diverse, high-performing feature subsets.
• The method offers a valuable tool for researchers and practitioners seeking robust feature selection strategies.