Scholarlink

<p>Feature selection is a crucial preprocessing phase in machine learning, aimed at identifying the most informative subset of features to improve classification accuracy while reducing computational overhead. Since feature selection is an NP-hard problem, metaheuristic algorithms have gained considerable attention as powerful wrapper-based optimization solutions. However, modern high-dimensional datasets with large feature spaces and relatively few samples pose substantial challenges, often resulting in diminished performance and excessive computational burdens. To address these limitations, this study proposes the Superior Predictive-based Human Memory Optimization (SPHMO) algorithm, an advanced extension of the Human Memory Optimization (HMO) method tailored for efficient and robust feature selection. SPHMO incorporates three integrated enhancement mechanisms: a Superior Point-Based Initialization Strategy combining Good Point Set theory with Oppositional-Based Learning to achieve well-distributed initial populations; a Predictive Trajectory-Based Motion Strategy (PTMS) that employs historical motion patterns to intelligently forecast search trajectories; and a Chaotic Vortex Perturbation (CVP) module that introduces dimension-wise probabilistic disturbances to escape local optima. The algorithm’s optimization performance is thoroughly validated using CEC2022 and CEC2017 benchmark suites, where it exhibits superior convergence behavior compared to classical, modern, and state-of-the-art competitors. Additionally, SPHMO’s feature selection capability is extensively evaluated across 24 diverse UCI datasets using multiple classifiers (k-NN, SVM, Decision Tree), and further demonstrated through real-world head and neck cancer biomarker identification using high-dimensional RNA-seq data. Experimental findings show that SPHMO delivers outstanding classification accuracy (81.65–100%) across various tasks while achieving extraordinary feature reduction, eliminating up to 99.97% of redundant attributes in genomic datasets. These results confirm SPHMO’s robustness, scalability, and practical suitability for addressing complex feature selection challenges in high-dimensional environments.</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 Superior Predictive Human Memory Optimization Algorithm for High-Dimensional Feature Selection for Head and Neck Cancer Biomarker Discovery

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • Abrar Yaqoob,
  • Khawaja T. Tasneem,
  • Prince Kumar Namdev,
  • G. V. V. Jagannadha Rao,
  • Mushtaq Ahmad Mir,
  • Nasreena Bashir,
  • R. Vijaya Lakshmi

摘要

Feature selection is a crucial preprocessing phase in machine learning, aimed at identifying the most informative subset of features to improve classification accuracy while reducing computational overhead. Since feature selection is an NP-hard problem, metaheuristic algorithms have gained considerable attention as powerful wrapper-based optimization solutions. However, modern high-dimensional datasets with large feature spaces and relatively few samples pose substantial challenges, often resulting in diminished performance and excessive computational burdens. To address these limitations, this study proposes the Superior Predictive-based Human Memory Optimization (SPHMO) algorithm, an advanced extension of the Human Memory Optimization (HMO) method tailored for efficient and robust feature selection. SPHMO incorporates three integrated enhancement mechanisms: a Superior Point-Based Initialization Strategy combining Good Point Set theory with Oppositional-Based Learning to achieve well-distributed initial populations; a Predictive Trajectory-Based Motion Strategy (PTMS) that employs historical motion patterns to intelligently forecast search trajectories; and a Chaotic Vortex Perturbation (CVP) module that introduces dimension-wise probabilistic disturbances to escape local optima. The algorithm’s optimization performance is thoroughly validated using CEC2022 and CEC2017 benchmark suites, where it exhibits superior convergence behavior compared to classical, modern, and state-of-the-art competitors. Additionally, SPHMO’s feature selection capability is extensively evaluated across 24 diverse UCI datasets using multiple classifiers (k-NN, SVM, Decision Tree), and further demonstrated through real-world head and neck cancer biomarker identification using high-dimensional RNA-seq data. Experimental findings show that SPHMO delivers outstanding classification accuracy (81.65–100%) across various tasks while achieving extraordinary feature reduction, eliminating up to 99.97% of redundant attributes in genomic datasets. These results confirm SPHMO’s robustness, scalability, and practical suitability for addressing complex feature selection challenges in high-dimensional environments.