Scholarlink

<p>In the realm of complex optimization problems, Meta-Heuristic Algorithms (MHAs), particularly swarm-based methods, have proven to be effective wrappers for Feature Selection (FS) problems. However, many such algorithms suffer from limited swarm diversity in later stages, leading to convergence on local optima. This study proposes a novel Hybrid Binary Bat–Sparrow Search Algorithm (HBBSSA) to address these limitations by combining strengths of the Sparrow Search Algorithm (SSA) and Bat Algorithm (BA) to enhance swarm diversity and exploration–exploitation balance. HBBSSA integrates multiple key strategies: (i) redefining out-of-bounds solutions; (ii) merging update mechanisms from both SSA and BA; iii) proposing a unique approach for refining loudness and pulse emission rate in BA. Additionally, to address high-dimensional data challenges, HBBSSA and its peers are combined with Mutual Information (MI) theory and tested on ten high-dimensional biological datasets (feature range: 4026–12600). Among all methods tested, HBBSSA achieves the most substantial improvements over the baseline <i>k</i>-Nearest Neighbor (<i>k</i>-NN) classifier, with an overall accuracy increase of 15.92% and an overall reduction of 71.96% in feature selection ratio across both low- and high-dimensional datasets. These findings are further validated by Wilcoxon signed-rank test, which confirms HBBSSA’s statistical significance, with success rates of 88.38% and 98.18% over competing methods on low- and high-dimensional datasets, respectively. HBBSSA is further evaluated against multiple cutting-edge FS methods recently proposed in the field. Overall, HBBSSA delivers exceptional performance and convergence, establishing itself as a robust optimization tool, particularly for large-scale biological data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploration–exploitation-balanced swarm intelligence with mutual information for dimensionality-independent feature selection and medical diagnosis

  • 期刊论文

  • 发表日期：

    2026年5月23日

查看全文
  • LiYun Jia,
  • DanYing Hu,
  • Tao Wang,
  • Amr A. Abohany,
  • Ahmed G. Gad

摘要

In the realm of complex optimization problems, Meta-Heuristic Algorithms (MHAs), particularly swarm-based methods, have proven to be effective wrappers for Feature Selection (FS) problems. However, many such algorithms suffer from limited swarm diversity in later stages, leading to convergence on local optima. This study proposes a novel Hybrid Binary Bat–Sparrow Search Algorithm (HBBSSA) to address these limitations by combining strengths of the Sparrow Search Algorithm (SSA) and Bat Algorithm (BA) to enhance swarm diversity and exploration–exploitation balance. HBBSSA integrates multiple key strategies: (i) redefining out-of-bounds solutions; (ii) merging update mechanisms from both SSA and BA; iii) proposing a unique approach for refining loudness and pulse emission rate in BA. Additionally, to address high-dimensional data challenges, HBBSSA and its peers are combined with Mutual Information (MI) theory and tested on ten high-dimensional biological datasets (feature range: 4026–12600). Among all methods tested, HBBSSA achieves the most substantial improvements over the baseline k-Nearest Neighbor (k-NN) classifier, with an overall accuracy increase of 15.92% and an overall reduction of 71.96% in feature selection ratio across both low- and high-dimensional datasets. These findings are further validated by Wilcoxon signed-rank test, which confirms HBBSSA’s statistical significance, with success rates of 88.38% and 98.18% over competing methods on low- and high-dimensional datasets, respectively. HBBSSA is further evaluated against multiple cutting-edge FS methods recently proposed in the field. Overall, HBBSSA delivers exceptional performance and convergence, establishing itself as a robust optimization tool, particularly for large-scale biological data.