In the context of increasing data scale, contemporary optimization algorithms struggle with cost and complexity in addressing the feature selection (FS) problem. This paper introduces a Harris hawks optimization (HHO) variant, enhanced with a multi-strategy augmentation (CXSHHO), for FS. The CXSHHO incorporates a communication and collaboration strategy (CC) into the baseline HHO, facilitating better information exchange among individuals, thereby expediting algorithmic convergence. Additionally, a directional crossover (DX) component refines the algorithm's ability to thoroughly explore the feature space. Furthermore, the soft-rime strategy (SR) broadens population diversity, enabling stochastic exploration of an extensive decision space and reducing the risk of local optima entrapment. The CXSHHO's global optimization efficacy is demonstrated through experiments on 30 functions from CEC2017, where it outperforms 15 established algorithms. Moreover, the paper presents a novel FS method based on CXSHHO, validated across 18 varied datasets from UCI. The results confirm CXSHHO's effectiveness in identifying subsets of features conducive to classification tasks.