A Novel Elite-Guided Hybrid Metaheuristic Algorithm for Efficient Feature Selection.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Feature selection is critical for improving machine learning model accuracy by identifying relevant features.
• The vast search space in feature selection necessitates meta-heuristic algorithms for efficient optimization.
• Existing optimization algorithms may face challenges like premature convergence and limited global exploration.

Purpose of the Study:

• To propose an improved meta-heuristic algorithm, Elite-guided Hybrid Northern Goshawk Optimization (EH-NGO), for enhanced feature selection.
• To improve the global optimization capabilities and convergence speed of the Northern Goshawk Optimization (NGO) algorithm.
• To develop and validate a novel feature selection method utilizing the proposed EH-NGO algorithm.

Main Methods:

• An elite-guided strategy is integrated into the NGO framework to direct population evolution.
• A vertical crossover mutation strategy is employed to enhance population diversity and global exploration.
• A boundary control strategy based on the global best solution is introduced to accelerate convergence.

Main Results:

• EH-NGO demonstrated superior global optimization performance on 30 benchmark functions (CEC2017 and CEC2022), outperforming eight state-of-the-art algorithms.
• The proposed feature selection method using EH-NGO was validated on 22 datasets of varying scales.
• Experimental results confirmed the method's effectiveness in selecting feature subsets that improve classification performance.

Conclusions:

• The EH-NGO algorithm offers significant improvements in global optimization and convergence speed.
• The novel feature selection method based on EH-NGO effectively identifies optimal feature subsets, leading to enhanced machine learning model performance.
• EH-NGO presents a promising approach for addressing complex feature selection challenges in machine learning.