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

High-dimensional small-sample data are widely encountered in medical diagnosis, bioinformatics, and industrial inspection, where traditional feature selection methods often suffer from premature convergence and local optima. To address these issues, this paper proposes a Hybrid Breeding-based Co-evolutionary Ant Colony Optimization method (HBACO) for feature selection. Inspired by the principle of hybrid breeding, in which individuals with distinct traits produce superior offspring through cross recombination, inheritance of desirable genes and continuous evolution, the proposed algorithm establishes a three-population collaborative framework. It consists of an ACO-based search population, an HRO-based evolutionary population and a cooperative feedback population that evolve iteratively together. Furthermore, we devise a heuristic strategy integrating correlation and genetic characteristics to help mine high-value feature subsets. Meanwhile, a collaborative pheromone updating mechanism is adopted to realize efficient knowledge sharing among populations. Experiments conducted on 13 high-dimensional datasets, including Colon and Lung, demonstrate that HBACO achieves superior classification accuracy, feature reduction performance, and convergence behavior compared with 10 representative algorithms. Specifically, HBACO improves the average classification accuracy by 3.9% and achieves an average feature dimensionality reduction rate of 91.4%. Statistical tests further confirm the significance of the proposed method. The results indicate that HBACO provides an effective and robust solution for high-dimensional feature selection problems.