Scholarlink

<p>Feature selection is a critical step in managing high-dimensional datasets, enhancing model performance, and improving interpretability. Traditional methods like Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Grey Wolf Optimizer (GWO) require careful parameter tuning and may stagnate or converge early in high dimensions. They also carry nontrivial runtime on large feature sets and may be sensitive to initialization. The Zebra Optimization Algorithm (ZO) has emerged as a promising tool for efficient feature selection; however, its comparative efficacy across various ZO variants remains underexplored. This study aims to evaluate and compare the performance of different ZO algorithm strategies, Original ZO (O-ZO), Adaptive Exploration–Exploitation ZO (AEE-ZO), Random Neighborhood Mutation ZO (RNM-ZO), Hybrid Inertia Control ZO (HIC-ZO), and Parameter-Free ZO (PF-ZO), in selecting optimal feature subsets across diverse high-dimensional datasets. A comprehensive experimental framework was established using 18 datasets spanning Biology, Health and Medicine, and Text domains. Each ZO variant was assessed based on multiple performance metrics, including classification accuracy, fitness optimization, feature reduction efficiency, Fisher Index, and execution time. Comparative analyses were conducted against traditional optimization algorithms, namely GA, PSO, and GWO. Across datasets, advanced ZO variants such as HIC-ZO and AEE-ZO performed competitively with the traditional baselines and often ranked among the top methods on several metrics. AEE-ZO balanced accuracy and feature reduction, while PF-ZO trailed on some datasets. Statistical analyses, including the Nemenyi post-hoc test, confirmed the significant superiority of advanced ZO variants over traditional methods in optimizing feature selection. The findings highlight the enhanced efficacy of hybrid and adaptive ZO algorithm variants in feature selection tasks within high-dimensional contexts. HIC-ZO and AEE-ZO, with their balanced approach to exploration and exploitation, offer significant improvements in classification performance and feature reduction efficiency. These advanced ZO variants present valuable alternatives to traditional optimization methods, fostering more efficient and interpretable models in complex data-driven applications.</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

Improved Zebra optimization strategies for efficient feature selection in high-dimensional spaces

  • 期刊论文

  • 发表日期：

    2025年9月12日

查看全文
  • Thompson Stephan,
  • S. Punitha,
  • Vinaytosh Mishra

摘要

Feature selection is a critical step in managing high-dimensional datasets, enhancing model performance, and improving interpretability. Traditional methods like Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Grey Wolf Optimizer (GWO) require careful parameter tuning and may stagnate or converge early in high dimensions. They also carry nontrivial runtime on large feature sets and may be sensitive to initialization. The Zebra Optimization Algorithm (ZO) has emerged as a promising tool for efficient feature selection; however, its comparative efficacy across various ZO variants remains underexplored. This study aims to evaluate and compare the performance of different ZO algorithm strategies, Original ZO (O-ZO), Adaptive Exploration–Exploitation ZO (AEE-ZO), Random Neighborhood Mutation ZO (RNM-ZO), Hybrid Inertia Control ZO (HIC-ZO), and Parameter-Free ZO (PF-ZO), in selecting optimal feature subsets across diverse high-dimensional datasets. A comprehensive experimental framework was established using 18 datasets spanning Biology, Health and Medicine, and Text domains. Each ZO variant was assessed based on multiple performance metrics, including classification accuracy, fitness optimization, feature reduction efficiency, Fisher Index, and execution time. Comparative analyses were conducted against traditional optimization algorithms, namely GA, PSO, and GWO. Across datasets, advanced ZO variants such as HIC-ZO and AEE-ZO performed competitively with the traditional baselines and often ranked among the top methods on several metrics. AEE-ZO balanced accuracy and feature reduction, while PF-ZO trailed on some datasets. Statistical analyses, including the Nemenyi post-hoc test, confirmed the significant superiority of advanced ZO variants over traditional methods in optimizing feature selection. The findings highlight the enhanced efficacy of hybrid and adaptive ZO algorithm variants in feature selection tasks within high-dimensional contexts. HIC-ZO and AEE-ZO, with their balanced approach to exploration and exploitation, offer significant improvements in classification performance and feature reduction efficiency. These advanced ZO variants present valuable alternatives to traditional optimization methods, fostering more efficient and interpretable models in complex data-driven applications.