Compound strategy based binary willow catkin optimization for feature selection

Hongming Chen, Qingli Yu, Guocai Yang, Hui Zhang, Gang Li, Qian Chen, Yonghua Li

Abstract: Feature selection is a crucial preprocessing technique that enhances the efficiency and accuracy of machine learning models by removing irrelevant and redundant features, thus reducing computational complexity and storage costs. However, existing binary swarm intelligence algorithms often encounter challenges such as getting trapped in local optima and lacking sufficient convergence performance. This study proposes a Compound Binary Willow Catkin Optimization (CBWCO) algorithm specifically designed for feature selection tasks. This paper extends the standard Willow Catkin Optimization (WCO) to its binary form, integrating a VU-shaped transfer function and a compound mutation strategy to improve adaptability for both high- and low-dimensional data while maintaining population diversity and strengthening global search capability. Experimental results on twelve benchmark datasets with different dimensions and fields from the UCI open-source databases show that CBWCO outperforms BWCO(S-shaped), Binary Grey Wolf Optimization (BGWO), Binary Particle Swarm Optimization (BPSO), Binary Differential Evolution and Genetic Algorithm(GA) in terms of convergence speed and classification performance across most datasets. These results highlight CBWCO’s potential advantages in feature selection and lay the groundwork for its future applications in fields such as signal recognition, fault diagnosis, and medical rehabilitation.

External IDs:dblp:journals/cluster/ChenYYZLCL25