Scholarlink

<p>Performing Feature Selection (FS) on high-dimensional medical data helps improve the quality of healthcare services and advances precision medicine. High-dimensional medical data often face the “curse of dimensionality,” with complex linear and nonlinear relationships among features that pose significant challenges for traditional FS algorithms. Metaheuristic Algorithms (MAs) have shown excellent performance in solving FS problems, but their tendency to get trapped in local optima can lead to the selection of irrelevant or redundant features in high-dimensional data. To address this, a Trajectory Prediction Strategy (TPS) is proposed, which calculates past optimization curves and innovatively uses them to predict search trajectories, enhancing the algorithm’s convergence performance. Additionally, a Direction Separation Strategy (DSS) was designed to implicitly reduce particle diversity loss caused by trajectory prediction and to guide individual mutations. The improved Harris Hawks Optimization (TDHHO), which incorporates the aforementioned strategies, achieves a better balance between convergence performance and diversity. Building on this foundation, the binary version of TDHHO, named BTDHHO, is designed using a V-shaped transfer function for FS in medical data. Experimental results demonstrate that the proposed binary BTDHHO algorithm achieves significantly higher accuracy compared to other binary algorithms across 9 medical datasets with over 5000 dimensions, indicating its potential as a powerful tool for FS in high-dimensional medical data.</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

Trajectory prediction and direction separation-based Harris hawks optimization for feature selection in high-dimensional medical data

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Zhilin Wang,
  • Lizhi Shao,
  • Mingjing Wang,
  • Huiling Chen

摘要

Performing Feature Selection (FS) on high-dimensional medical data helps improve the quality of healthcare services and advances precision medicine. High-dimensional medical data often face the “curse of dimensionality,” with complex linear and nonlinear relationships among features that pose significant challenges for traditional FS algorithms. Metaheuristic Algorithms (MAs) have shown excellent performance in solving FS problems, but their tendency to get trapped in local optima can lead to the selection of irrelevant or redundant features in high-dimensional data. To address this, a Trajectory Prediction Strategy (TPS) is proposed, which calculates past optimization curves and innovatively uses them to predict search trajectories, enhancing the algorithm’s convergence performance. Additionally, a Direction Separation Strategy (DSS) was designed to implicitly reduce particle diversity loss caused by trajectory prediction and to guide individual mutations. The improved Harris Hawks Optimization (TDHHO), which incorporates the aforementioned strategies, achieves a better balance between convergence performance and diversity. Building on this foundation, the binary version of TDHHO, named BTDHHO, is designed using a V-shaped transfer function for FS in medical data. Experimental results demonstrate that the proposed binary BTDHHO algorithm achieves significantly higher accuracy compared to other binary algorithms across 9 medical datasets with over 5000 dimensions, indicating its potential as a powerful tool for FS in high-dimensional medical data.