BMSZOA: a Binary Zebra Optimization Algorithm with Multi-Strategy Improvement for Feature Selection Problems

Addressing the issues of strong parameter sensitivity, high risk of premature convergence, and low computational efficiency for high-dimensional data in existing binary optimization algorithms for feature selection, this paper proposes a multi-strategy improved binary zebra optimization algorithm. The algorithm achieves intelligent mapping from continuous space to discrete features through a dynamic S-shaped function, combines a hybrid exploration strategy of Lévy flight and Gaussian perturbation to balance global search and local exploitation capabilities, and monitors population diversity based on information entropy to prevent premature convergence. Experiments on 18 standard datasets demonstrate that the algorithm achieves an average classification accuracy of $\mathbf{9 6. 2 1 \%}$, outperforming six mainstream algorithms by 1.34 %, with the highest accuracy ranking on 16 datasets; the average feature selection rate is reduced to 34.7 %, requiring only 33.1 % of features to achieve 100 % accuracy on the Wine dataset; convergence speed is improved by 32 %, and stability on noisy data is enhanced by over 40 %. This study provides an efficient and robust solution for high-dimensional feature selection, with potential applications in biomedical diagnosis and financial risk control.