Scholarlink

<p>In recent years, the salp swarm algorithm (SSA) has become an efficient tool for feature selection (FS) problems. However, the algorithm has drawbacks, including local optimality and a limited convergence rate. Therefore, we propose NSSA, an enhanced version of SSA that integrates four strategies. Firstly, a multi-round voting mechanism based on three filter methods is presented to achieve a high-quality initial population. Furthermore, an adaptive Lagrange interpolation inertia weight is introduced to promote the adaptative capability by defining a testing phase. Additionally, the position updates mathematical models of leader and followers are modified by taking advantage of more important positions to enhance the search performance. Finally, a novel elite generalized opposition-based learning is presented to accelerate the convergence rate. The NSSA is compared with 7 metaheuristic algorithms on 16 benchmark datasets. The outcomes indicate that the NSSA achieves better fitness values and smaller feature sizes on the majority of datasets.</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 feature selection method based on salp swarm algorithm with a multi-round voting mechanism

  • 期刊论文

  • 发表日期：

    2025年8月13日

查看全文
  • Hongbo Zhang,
  • Haohuan Nan,
  • Xiaofeng Yue,
  • Xueliang Gao

摘要

In recent years, the salp swarm algorithm (SSA) has become an efficient tool for feature selection (FS) problems. However, the algorithm has drawbacks, including local optimality and a limited convergence rate. Therefore, we propose NSSA, an enhanced version of SSA that integrates four strategies. Firstly, a multi-round voting mechanism based on three filter methods is presented to achieve a high-quality initial population. Furthermore, an adaptive Lagrange interpolation inertia weight is introduced to promote the adaptative capability by defining a testing phase. Additionally, the position updates mathematical models of leader and followers are modified by taking advantage of more important positions to enhance the search performance. Finally, a novel elite generalized opposition-based learning is presented to accelerate the convergence rate. The NSSA is compared with 7 metaheuristic algorithms on 16 benchmark datasets. The outcomes indicate that the NSSA achieves better fitness values and smaller feature sizes on the majority of datasets.