<p>In the field of medicine, there exists a vast amount of biomedical and clinical data, with a majority of which being high-dimensional gene expression data. As the dimension of genetic data increases, the complexity and computational cost of processing also increase significantly. How to select effective features is an important and challenging research topic. To solve this problem, this work proposed a new feature selection algorithm, named SCHGS, which integrates crisscross selection (CC) and positive greedy selection mechanisms (PGS) into the Hunger Games Search (HGS) algorithm. CC effectively enhances the global optimization ability of the SCHGS, and PGS can speed up the convergence of the SCHGS. The experimental results indicate that, compared to HGS, SCHGS achieves a better balance between exploration and exploitation, enhances the ability to escape local optima, and demonstrates greater robustness on both low-dimensional and high-dimensional problems. The comparative experimental results indicate that SCHGS achieves significantly faster convergence and more effectively avoids and overcomes local optima than other metaheuristic algorithms. The binary version of SCHGS (BSCHGS) was applied to feature selection tasks on six high-dimensional gene datasets, where it outperformed six other optimization algorithms and four conventional feature selection methods in classification accuracy. Notably, on the Lung_Cancer dataset with 12,601 features, BSCHGS achieved an average classification error as low as 0.0143, while on the DLBCL and SRBCT datasets, the average error dropped to 0, demonstrating outstanding performance. SCHGS is thus proven to be a robust and efficient optimization algorithm that shows significant advantages in high-dimensional feature selection tasks.</p>