Scholarlink

<p>Feature selection (FS) is a critical preprocessing step in data mining and machine learning. The Victoria Amazonica Optimization (VAO) algorithm is a bio-inspired metaheuristic that mimics the growth and competition of the giant Amazon water lilies to solve optimization problems. We propose an enhanced VAO algorithm (SMu3-VAO) incorporating competitive selection strategies and dynamic mutation factors to improve global function optimization and feature selection. Two novel selection strategies, equal interval extraction and niche competition—are introduced to boost population diversity during fitness evaluation. Additionally, we replace the original linear decay mutation factor with an adaptive dynamic adjustment mechanism, enhancing global search capability and convergence. Extensive experiments on CEC 2022 test functions confirm SMu3-VAO’s superior accuracy and faster convergence. For feature selection, SMu3-VAO outperforms competing algorithms across 18 UCI datasets, as validated by KNN classifier performance. The experimental results show that the SMu3-VAO algorithm demonstrates significant advantages in key metrics such as classification accuracy, mean fitness value and standard deviation when solving the FS problems. The SMu3-VAO algorithm improves the average accuracy by 5.7% and the average number of selected features by 46.4% compared to the original algorithm.</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

Victoria Amazonica optimization algorithm based on competitive selection strategy and dynamic mutation factor to solve feature selection problems

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • Xin-Ru Ma,
  • Jie-Sheng Wang,
  • Yong-Cheng Sun,
  • Yu-Cai Wang,
  • Yu-Wei Song,
  • Yu-Liang Qi

摘要

Feature selection (FS) is a critical preprocessing step in data mining and machine learning. The Victoria Amazonica Optimization (VAO) algorithm is a bio-inspired metaheuristic that mimics the growth and competition of the giant Amazon water lilies to solve optimization problems. We propose an enhanced VAO algorithm (SMu3-VAO) incorporating competitive selection strategies and dynamic mutation factors to improve global function optimization and feature selection. Two novel selection strategies, equal interval extraction and niche competition—are introduced to boost population diversity during fitness evaluation. Additionally, we replace the original linear decay mutation factor with an adaptive dynamic adjustment mechanism, enhancing global search capability and convergence. Extensive experiments on CEC 2022 test functions confirm SMu3-VAO’s superior accuracy and faster convergence. For feature selection, SMu3-VAO outperforms competing algorithms across 18 UCI datasets, as validated by KNN classifier performance. The experimental results show that the SMu3-VAO algorithm demonstrates significant advantages in key metrics such as classification accuracy, mean fitness value and standard deviation when solving the FS problems. The SMu3-VAO algorithm improves the average accuracy by 5.7% and the average number of selected features by 46.4% compared to the original algorithm.