A Fast Hybrid Feature Selection Based on Correlation-Guided Clustering and Particle Swarm Optimization for...

Area of Science:

• Machine Learning
• Data Mining
• Computational Intelligence

Background:

• High-dimensional data presents challenges like the "curse of dimensionality" and high computational costs, limiting the application of evolutionary algorithms in feature selection (FS).
• Existing FS methods often struggle to balance search space reduction with the identification of optimal feature subsets efficiently.

Purpose of the Study:

• To propose a novel three-phase hybrid FS algorithm (HFS-C-P) that addresses both the curse of dimensionality and high computational costs.
• To integrate filter, feature clustering, and evolutionary algorithms to create an efficient and effective FS solution.

Main Methods:

• A three-phase hybrid FS algorithm (HFS-C-P) combining filter methods, correlation-guided clustering, and particle swarm optimization (PSO).
• Phase 1 & 2: Employ filter FS and feature clustering for search space reduction.
• Phase 3: Utilize an evolutionary algorithm with global searchability for optimal subset identification, enhanced by symmetric uncertainty and improved integer PSO.

Main Results:

• The proposed HFS-C-P algorithm was validated on 18 real-world datasets against nine existing FS algorithms.
• Experimental results demonstrated that HFS-C-P successfully obtains a high-quality feature subset.
• The algorithm achieved the lowest computational cost among the compared methods.

Conclusions:

• The HFS-C-P algorithm effectively tackles the curse of dimensionality and high computational costs in high-dimensional FS.
• The hybrid approach integrates diverse FS strategies to achieve superior performance and efficiency.
• The proposed method offers a promising solution for efficient and effective feature selection in complex datasets.