High-Dimensional Feature Selection Using Improved Hybrid Breeding Optimization Algorithm with Feature Grouping.

Area of Science:

• Biomedical data analysis
• Machine learning
• Optimization algorithms

Background:

• High-dimensional biomedical data presents challenges for classification.
• Conventional metaheuristic algorithms struggle with premature convergence and reduced population diversity.
• Effective feature selection is crucial for improving classification performance.

Purpose of the Study:

• To propose a Feature Grouping and Improved Hybrid Breeding Optimization (FGIHBO) framework.
• To address the limitations of conventional metaheuristic algorithms in high-dimensional biomedical data.
• To enhance classification accuracy through improved feature selection.

Main Methods:

• Hierarchical partitioning of feature space using Maximum Relevance Minimum Redundancy (MRMR) and Symmetric Uncertainty (SU).
• Development of a Multi-Strategy Synergistic Improved Hybrid Breeding Optimization (MSIHBO) algorithm.
• Incorporation of Grey Wolf Optimizer (GWO) guidance and Shannon entropy-adaptive simulated annealing for balanced exploration and exploitation.

Main Results:

• MSIHBO demonstrated robust optimization performance on the CEC2022 benchmark suite.
• FGIHBO achieved average classification accuracies of 92.77%–97.66% on eleven high-dimensional biomedical datasets.
• The proposed framework improved average classification accuracy by 1.47%–27.46% compared to existing algorithms.

Conclusions:

• FGIHBO is effective and robust for feature selection in high-dimensional biomedical data.
• The framework demonstrates significant improvements in classification accuracy.
• The scalability of FGIHBO is confirmed for complex biomedical datasets.