Scholarlink

<p>Addressing the persistent challenge of balancing global exploration and local exploitation in complex optimization and feature selection tasks, we introduce Hybrid Crayfish Optimization Algorithm with Harmonic Search (HCOAHS), an adaptive hybrid metaheuristic that synergistically integrates swarm intelligence mechanisms with memory-driven refinement strategies. The framework implements temperature-gated phase-specific integration, where Harmonic Search (HS) refinement activates exclusively during exploitation phases (temperature <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\le 30^{\circ }\text {C}\)</EquationSource> </InlineEquation>), ensuring efficient resource allocation between exploration and exploitation. A distinctive innovation involves population-seeded harmony memory initialization, creating bidirectional information flow that accelerates convergence while maintaining diversity. Comprehensive evaluation on CEC 2014, 2017, 2020, and 2022 benchmark suites demonstrates HCOAHS’s superiority, achieving average ranks of 1.97, 1.58, 1.7, and 1.58, respectively, across 30-120 functions, significantly outperforming ten state-of-the-art algorithms, including COA, QIO, MGO, and advanced DE variants (L-SHADE, SADE). Wilcoxon rank-sum tests confirm statistical significance (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(p&lt;0.05\)</EquationSource> </InlineEquation>) across all comparisons except QIO (<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(p=0.092\)</EquationSource> </InlineEquation>). The binary variant addresses feature selection on 20 real-world datasets, achieving 60-70% dimensionality reduction while improving classification accuracy by 2-5% with K-NN classifiers. Statistical validation at 5% significance level establishes HCOAHS’s robust performance across biology, politics, physics, and computer science domains. Complexity analysis reveals <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\mathcal {O}(T \cdot N \cdot (D+F_{\text {obj}}))\)</EquationSource> </InlineEquation> computational cost with mean runtime ranging 1.87-2.41s on standard benchmarks. Sensitivity analysis confirms stability across <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\text {HMCR} \in [0.85,0.95]\)</EquationSource> </InlineEquation>, <InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(\text {PAR} \in [0.05,0.15]\)</EquationSource> </InlineEquation>, and <InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(\text {HMS} \in [20, 30]\)</EquationSource> </InlineEquation>, demonstrating parameter robustness. These findings position HCOAHS as a powerful optimization framework with strong generalization capabilities across continuous and discrete domains, offering practical applicability to high-dimensional data-driven problems.</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

Adaptive hybrid crayfish - Harmonic search metaheuristic for robust feature selection in supervised classification

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Binanda Maiti,
  • Saptadeep Biswas,
  • Madhujit Deb,
  • Uttam Kumar Bera

摘要

Addressing the persistent challenge of balancing global exploration and local exploitation in complex optimization and feature selection tasks, we introduce Hybrid Crayfish Optimization Algorithm with Harmonic Search (HCOAHS), an adaptive hybrid metaheuristic that synergistically integrates swarm intelligence mechanisms with memory-driven refinement strategies. The framework implements temperature-gated phase-specific integration, where Harmonic Search (HS) refinement activates exclusively during exploitation phases (temperature \(\le 30^{\circ }\text {C}\) ), ensuring efficient resource allocation between exploration and exploitation. A distinctive innovation involves population-seeded harmony memory initialization, creating bidirectional information flow that accelerates convergence while maintaining diversity. Comprehensive evaluation on CEC 2014, 2017, 2020, and 2022 benchmark suites demonstrates HCOAHS’s superiority, achieving average ranks of 1.97, 1.58, 1.7, and 1.58, respectively, across 30-120 functions, significantly outperforming ten state-of-the-art algorithms, including COA, QIO, MGO, and advanced DE variants (L-SHADE, SADE). Wilcoxon rank-sum tests confirm statistical significance ( \(p<0.05\) ) across all comparisons except QIO ( \(p=0.092\) ). The binary variant addresses feature selection on 20 real-world datasets, achieving 60-70% dimensionality reduction while improving classification accuracy by 2-5% with K-NN classifiers. Statistical validation at 5% significance level establishes HCOAHS’s robust performance across biology, politics, physics, and computer science domains. Complexity analysis reveals \(\mathcal {O}(T \cdot N \cdot (D+F_{\text {obj}}))\) computational cost with mean runtime ranging 1.87-2.41s on standard benchmarks. Sensitivity analysis confirms stability across \(\text {HMCR} \in [0.85,0.95]\) , \(\text {PAR} \in [0.05,0.15]\) , and \(\text {HMS} \in [20, 30]\) , demonstrating parameter robustness. These findings position HCOAHS as a powerful optimization framework with strong generalization capabilities across continuous and discrete domains, offering practical applicability to high-dimensional data-driven problems.