IHBOFS: A Biomimetics-Inspired Hybrid Breeding Optimization Algorithm for High-Dimensional Feature Selection.

Area of Science:

• Computational Intelligence
• Data Science
• Optimization Algorithms

Background:

• Effective data preprocessing is crucial due to explosive data growth.
• Evolutionary and swarm intelligence algorithms show promise in feature selection but struggle with large-scale problems.
• Premature convergence and limited exploration hinder performance in existing algorithms.

Purpose of the Study:

• To propose IHBOFS, a novel biomimetics-inspired optimization framework for enhanced feature selection.
• To address limitations of existing algorithms in large-scale, high-dimensional feature selection.
• To improve performance and stability in data preprocessing tasks.

Main Methods:

• Developed IHBOFS integrating Good Point Set and Elite Opposition-Based Learning for diverse initialization.
• Implemented adaptive exploitation-exploration balancing strategies for subpopulations to mitigate premature convergence.
• Extended IHBOFS with continuous-to-discrete mapping for discrete feature selection problems.

Main Results:

• Ablation studies on CEC2022 benchmark functions validated the effectiveness of proposed strategies.
• IHBOFS achieved an average classification accuracy of 92.57% on six real-world datasets.
• Comparative experiments demonstrated IHBOFS's superiority over nine metaheuristic methods, including HHO and ACO.

Conclusions:

• IHBOFS effectively enhances feature selection performance and stability in high-dimensional datasets.
• The integrated adaptive strategies successfully mitigate premature convergence and improve exploration.
• IHBOFS offers a robust and superior solution for complex feature selection tasks in data science.