Scholarlink

<p>Feature selection is a critical process for all machine learning models, as their performance relies heavily on the relevance of the selected features. It is a challenging task that typically is addressed with evolutionary algorithms. This paper presents the Chaotic and Adaptive Manta Ray Foraging Optimization (CAMRFO), which proposes a novel integrated framework to address the common metaheuristic challenges of premature convergence and poor exploration–exploitation balance. The framework's contribution lies in the synergistic integration of three strategies: a chaos-based mechanism to enhance global exploration, an adaptive t-distribution to maintain population diversity, and an Adaptive Control Parameter (ACP) strategy to dynamically manage the search process. The performance of the CAMRFO framework was evaluated against top optimization techniques for 20 benchmark functions, 14 UCI datasets, and 3 engineering problems. Results show that CAMRFO outperforms its counterparts on 18 of the 20 benchmark functions and on all three engineering problems. In the feature selection tasks, CAMRFO consistently achieved the highest classification accuracy on all 14 datasets, recording average accuracies of 93.6% with the KNN classifier and 98.2% with the Random Forest classifier. Furthermore, it demonstrated a superior balance between accuracy and feature reduction, improving the average fitness value by up to 3.1% compared to its closest competitor. These results confirm that CAMRFO is an extremely effective and robust optimization framework.</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

An improved manta ray foraging optimization for feature selection and engineering applications

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Yazdan Yadollahi,
  • Behnam Mohammad Hasani Zade,
  • Najme Mansouri

摘要

Feature selection is a critical process for all machine learning models, as their performance relies heavily on the relevance of the selected features. It is a challenging task that typically is addressed with evolutionary algorithms. This paper presents the Chaotic and Adaptive Manta Ray Foraging Optimization (CAMRFO), which proposes a novel integrated framework to address the common metaheuristic challenges of premature convergence and poor exploration–exploitation balance. The framework's contribution lies in the synergistic integration of three strategies: a chaos-based mechanism to enhance global exploration, an adaptive t-distribution to maintain population diversity, and an Adaptive Control Parameter (ACP) strategy to dynamically manage the search process. The performance of the CAMRFO framework was evaluated against top optimization techniques for 20 benchmark functions, 14 UCI datasets, and 3 engineering problems. Results show that CAMRFO outperforms its counterparts on 18 of the 20 benchmark functions and on all three engineering problems. In the feature selection tasks, CAMRFO consistently achieved the highest classification accuracy on all 14 datasets, recording average accuracies of 93.6% with the KNN classifier and 98.2% with the Random Forest classifier. Furthermore, it demonstrated a superior balance between accuracy and feature reduction, improving the average fitness value by up to 3.1% compared to its closest competitor. These results confirm that CAMRFO is an extremely effective and robust optimization framework.