High-Dimensional Small-Sample Feature Selection Using Co-Evolutionary Ant Colony Optimization Inspired by Heterosis.

Area of Science:

• Computational Biology
• Machine Learning
• Data Science

Background:

• High-dimensional small-sample data present challenges for traditional feature selection methods, often leading to premature convergence and local optima.
• Applications in medical diagnosis, bioinformatics, and industrial inspection necessitate robust feature selection techniques.

Purpose of the Study:

• To propose a Hybrid Breeding-based Co-evolutionary Ant Colony Optimization (HBACO) method for superior feature selection.
• To address the limitations of existing methods in handling high-dimensional small-sample data.

Main Methods:

• Developed a three-population collaborative framework: ACO-based search, HRO-based evolutionary, and cooperative feedback populations.
• Integrated a heuristic strategy combining correlation and genetic characteristics for high-value feature subset mining.
• Implemented a collaborative pheromone updating mechanism for efficient inter-population knowledge sharing.

Main Results:

• HBACO demonstrated superior classification accuracy, achieving an average improvement of 3.9%.
• The method achieved a significant average feature dimensionality reduction rate of 91.4%.
• Experimental results on 13 high-dimensional datasets showed improved performance and convergence behavior compared to 10 representative algorithms.

Conclusions:

• HBACO offers an effective and robust solution for feature selection in high-dimensional datasets.
• The proposed method overcomes the limitations of traditional algorithms, providing better accuracy and dimensionality reduction.
• Statistical tests confirmed the significance and reliability of the HBACO method.