Scholarlink

<p>Feature selection (FS) is a crucial process in machine learning and data mining that aims to reduce dimensionality while maintaining high classification accuracy. The presence of irrelevant and redundant features in high-dimensional datasets not only increases computational overhead but also degrades model generalization. Traditional FS techniques, including filter, wrapper, and embedded methods, often struggle with scalability and efficiency, especially when applied to large and complex datasets. Metaheuristic-based FS techniques, such as Genetic Algorithms (GA) and Particle Swarm Optimization (PSO), have been widely explored to address these challenges. However, these methods frequently suffer from premature convergence and suboptimal feature selection, leading to performance inconsistencies across diverse datasets. In response to these limitations, this study proposes a novel FS algorithm, Adaptive Quantum Artificial Flora Optimization (AQAFO), which integrates quantum-inspired crossover and mutation operations within the Artificial Flora Optimization (AFO) framework. AQAFO introduces adaptive population scaling based on dataset complexity, ensuring optimal trade-offs between search diversity and computational efficiency. The performance of AQAFO is evaluated on 21 benchmark datasets from the ASU Feature Selection Repository, utilizing four classifiers (KNN, LR, RF, and SVM) to assess its effectiveness. Comparative analysis against state-of-the-art FS methods, including BDE, BDMO, SSA, GMSMFO, GNDO, and FPGFS, demonstrates that AQAFO consistently achieves higher classification accuracy while selecting an optimal number of features. The results confirm that AQAFO significantly improves feature subset selection, enhances classification accuracy, and prevents premature convergence, making it a highly efficient FS approach. Furthermore, AQAFO exhibits strong generalization capabilities across various data domains, including biomedical, image processing, and high-dimensional text classification. The proposed algorithm ensures computational efficiency while reducing feature selection bias, establishing itself as a robust FS solution for high-dimensional machine learning 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

Adaptive Quantum Artificial Flora Optimization for Feature Selection in High-Dimensional Data

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Keerthi Gabbi Reddy,
  • Deepasikha Mishra

摘要

Feature selection (FS) is a crucial process in machine learning and data mining that aims to reduce dimensionality while maintaining high classification accuracy. The presence of irrelevant and redundant features in high-dimensional datasets not only increases computational overhead but also degrades model generalization. Traditional FS techniques, including filter, wrapper, and embedded methods, often struggle with scalability and efficiency, especially when applied to large and complex datasets. Metaheuristic-based FS techniques, such as Genetic Algorithms (GA) and Particle Swarm Optimization (PSO), have been widely explored to address these challenges. However, these methods frequently suffer from premature convergence and suboptimal feature selection, leading to performance inconsistencies across diverse datasets. In response to these limitations, this study proposes a novel FS algorithm, Adaptive Quantum Artificial Flora Optimization (AQAFO), which integrates quantum-inspired crossover and mutation operations within the Artificial Flora Optimization (AFO) framework. AQAFO introduces adaptive population scaling based on dataset complexity, ensuring optimal trade-offs between search diversity and computational efficiency. The performance of AQAFO is evaluated on 21 benchmark datasets from the ASU Feature Selection Repository, utilizing four classifiers (KNN, LR, RF, and SVM) to assess its effectiveness. Comparative analysis against state-of-the-art FS methods, including BDE, BDMO, SSA, GMSMFO, GNDO, and FPGFS, demonstrates that AQAFO consistently achieves higher classification accuracy while selecting an optimal number of features. The results confirm that AQAFO significantly improves feature subset selection, enhances classification accuracy, and prevents premature convergence, making it a highly efficient FS approach. Furthermore, AQAFO exhibits strong generalization capabilities across various data domains, including biomedical, image processing, and high-dimensional text classification. The proposed algorithm ensures computational efficiency while reducing feature selection bias, establishing itself as a robust FS solution for high-dimensional machine learning applications.