Scholarlink

<p>Feature selection, a critical technique for identifying optimal feature subsets, enhances model accuracy while reducing feature dimensionality. Although the recently proposed metaheuristic Horned Lizard Optimization Algorithm (HLOA) exhibits robust stochastic search capabilities for complex optimization problems, it is inherently incompatible with discrete binary tasks such as feature selection. While sigmoid functions conventionally bridge this gap, this paper introduces an innovative Euclidean-distance-based binarization mechanism and its enhanced variant to adapt HLOA's superior search performance to feature selection. Experimental validation across 20 UCI (University of California, Irvine) benchmark datasets demonstrates the efficacy of the proposed methods. Notably, on high-dimensional datasets (dimensionality &gt; 1,000), our algorithms achieve significant reductions in feature size while consistently improving predictive accuracy.</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

A Euclidean Distance-Based Novel Algorithm for Binary Feature Selection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月28日

查看全文
  • Haiyan Wang,
  • Kai Han,
  • Chuang Li,
  • Jian Zhao,
  • Na Che,
  • Xiaotong Liu

摘要

Feature selection, a critical technique for identifying optimal feature subsets, enhances model accuracy while reducing feature dimensionality. Although the recently proposed metaheuristic Horned Lizard Optimization Algorithm (HLOA) exhibits robust stochastic search capabilities for complex optimization problems, it is inherently incompatible with discrete binary tasks such as feature selection. While sigmoid functions conventionally bridge this gap, this paper introduces an innovative Euclidean-distance-based binarization mechanism and its enhanced variant to adapt HLOA's superior search performance to feature selection. Experimental validation across 20 UCI (University of California, Irvine) benchmark datasets demonstrates the efficacy of the proposed methods. Notably, on high-dimensional datasets (dimensionality > 1,000), our algorithms achieve significant reductions in feature size while consistently improving predictive accuracy.