Scholarlink

<p>To address the limitations of the original whale optimization algorithm (WOA) such as insufficient population diversity, poor balance between exploration and exploitation, and proneness to local optima, this study proposes a median-driven whale optimization algorithm (MWOA) for feature selection, with a focus on analyzing students’ reading habits. MWOA integrates four key improved strategies: Tent chaotic mapping enhances the diversity of the initial population. The nonlinear convergence factor dynamically balances global exploration and local exploitation. The position update strategy based on median divides the population into two groups and applies Gaussian mutation and Cauchy mutation, respectively, to improve local search and global efficiency. The differential mutation strategy based on median optimizes the current generation optimal solution to avoid local optima. The performance of MWOA is validated on 30-D CEC2017, 50-D CEC2017, 10-D CEC2022 and 20-D CEC2022 benchmark functions, where it outperforms 10 state-of-the-art optimization algorithms in terms of convergence speed, solution accuracy, and stability. Furthermore, MWOA is combined with Support Vector Machine (SVM) to form MWOA-SVM for feature selection tasks. Experimental results on 11 UCI datasets and a students’ reading habits dataset from KAGGLE platform show that MWOA-SVM achieves superior performance in worst fitness, best fitness, average fitness, classification accuracy, precision, recall, F1-score, and selected feature number. Particularly in predicting students’ reading habits, MWOA-SVM effectively filters redundant features, reduces model complexity, and provides data support for library resource optimization and reading guidance strategies. This paper demonstrates MWOA’s excellent performance in solving high-dimensional optimization and feature selection problems, with broad application prospects in educational data mining and real-world intelligent optimization scenarios. For large-scale feature search in students’ reading habits and high-dimensional UCI datasets, MWOA’s multi-strategy fusion increases computational complexity. Relying on supercomputing’s parallel and distributed capabilities, MWOA-SVM enables fast search in high-dimensional spaces, supporting large-scale educational data mining.</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

Median-driven whale optimization algorithm with differential mutation for feature selection in students’ reading habits

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Yige Xue,
  • Fa Sun

摘要

To address the limitations of the original whale optimization algorithm (WOA) such as insufficient population diversity, poor balance between exploration and exploitation, and proneness to local optima, this study proposes a median-driven whale optimization algorithm (MWOA) for feature selection, with a focus on analyzing students’ reading habits. MWOA integrates four key improved strategies: Tent chaotic mapping enhances the diversity of the initial population. The nonlinear convergence factor dynamically balances global exploration and local exploitation. The position update strategy based on median divides the population into two groups and applies Gaussian mutation and Cauchy mutation, respectively, to improve local search and global efficiency. The differential mutation strategy based on median optimizes the current generation optimal solution to avoid local optima. The performance of MWOA is validated on 30-D CEC2017, 50-D CEC2017, 10-D CEC2022 and 20-D CEC2022 benchmark functions, where it outperforms 10 state-of-the-art optimization algorithms in terms of convergence speed, solution accuracy, and stability. Furthermore, MWOA is combined with Support Vector Machine (SVM) to form MWOA-SVM for feature selection tasks. Experimental results on 11 UCI datasets and a students’ reading habits dataset from KAGGLE platform show that MWOA-SVM achieves superior performance in worst fitness, best fitness, average fitness, classification accuracy, precision, recall, F1-score, and selected feature number. Particularly in predicting students’ reading habits, MWOA-SVM effectively filters redundant features, reduces model complexity, and provides data support for library resource optimization and reading guidance strategies. This paper demonstrates MWOA’s excellent performance in solving high-dimensional optimization and feature selection problems, with broad application prospects in educational data mining and real-world intelligent optimization scenarios. For large-scale feature search in students’ reading habits and high-dimensional UCI datasets, MWOA’s multi-strategy fusion increases computational complexity. Relying on supercomputing’s parallel and distributed capabilities, MWOA-SVM enables fast search in high-dimensional spaces, supporting large-scale educational data mining.