An improved binary arithmetic optimization algorithm with synchronous strong guidance for high-dimensional feature selection

To address the high redundancy and excessive dimensional issues of high-dimensional data, an improved binary arithmetic optimization algorithm with synchronous strong guidance (SSG-BSIAOA) is proposed. Firstly, to increase the diversity of exploration and the richness of decision-making directions, a bidirectional probability selection strategy is proposed. Then, a trigonometric function factor strategy with eight trigonometric function factors are respectively used to replace the fixed constants so as to increase the algorithm diversity. To improve the insufficient optimization ability of the wrapper method, a synchronous strong guidance strategy based on mutual information and ReliefF dual indicators is proposed, which explores high-quality candidate features from multiple perspectives and breaks through the limitations of single-index evaluation. In the experimental part, the effectiveness of the bidirectional probability selection strategy and the trigonometric function factor strategy is verified through CEC-2022 benchmark functions, and it is proved that the sec-IAOA algorithm is the best variant. Then the performance of the SSG-BSIAOA is verified through 21 high-dimensional UCI datasets, proving the advantages of the proposed method in fitness, accuracy and dimension compression ability. Then, SSG-BSIAOA is applied to 18 micro-array datasets. By comparing with 9 algorithms, it is found that the SSG-BSIAOA has strong optimization ability, good stability, high accuracy and feature simplification. Finally, non-parametric statistical tests, including direct paired Wilcoxon signed-rank comparisons with Holm correction, were conducted to further assess the robustness of the observed performance differences.