Scholarlink

<p>Effective feature selection becomes increasingly important as medical data volumes increase. Despite this, existing dimensionality reduction techniques typically fail to address the unique challenges posed by medical datasets, such as their high dimensionality, class imbalance, and ethical imperative to use sensitive patient information as little as possible. To improve feature selection for medical data, this paper presents a novel reinforcement learning approach. Unlike traditional wrapper methods, which are prone to local optima and premature convergence, our approach leverages a chaos-based upper confidence bound (C-UCB) algorithm to dynamically adjust key parameters, such as the exploration coefficient (<i>c</i>) and the number of inner loop iterations (<i>N</i>), based on the mean reward from previous iterations. This dynamic adjustment not only improves algorithm performance but also significantly reduces computation time. Our experiments demonstrate that the proposed method outperforms state-of-the-art algorithms, improving the objective function by 51.43%, 42.86%, 88.57%, 71.43%, and 68.57%, compared to simulated annealing (SA), gray wolf optimization (GWO), pathfinder algorithm (PFA), <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10115_2025_2427_Article_IEq1.gif" Format="GIF" Height="10" Rendition="HTML" Resolution="72" Type="Linedraw" Width="11" /> </InlineMediaObject> <EquationSource Format="TEX">\(\varepsilon \)</EquationSource> <EquationSource Format="MATHML"><math> <mi>ε</mi> </math></EquationSource> </InlineEquation>-Greedy, and UCB, respectively. This innovative approach has the potential to significantly enhance the accuracy and efficiency of feature selection in medical applications, enabling the development of more reliable and interpretable predictive models while minimizing the use of sensitive patient 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

C-UCB: chaos upper confidence bound in reinforcement learning for feature selection

  • 期刊论文

  • 发表日期：

    2025年4月19日

查看全文
  • Aboozar Zandvakili,
  • Mohammad Masoud Javidi,
  • Najme Mansouri

摘要

Effective feature selection becomes increasingly important as medical data volumes increase. Despite this, existing dimensionality reduction techniques typically fail to address the unique challenges posed by medical datasets, such as their high dimensionality, class imbalance, and ethical imperative to use sensitive patient information as little as possible. To improve feature selection for medical data, this paper presents a novel reinforcement learning approach. Unlike traditional wrapper methods, which are prone to local optima and premature convergence, our approach leverages a chaos-based upper confidence bound (C-UCB) algorithm to dynamically adjust key parameters, such as the exploration coefficient (c) and the number of inner loop iterations (N), based on the mean reward from previous iterations. This dynamic adjustment not only improves algorithm performance but also significantly reduces computation time. Our experiments demonstrate that the proposed method outperforms state-of-the-art algorithms, improving the objective function by 51.43%, 42.86%, 88.57%, 71.43%, and 68.57%, compared to simulated annealing (SA), gray wolf optimization (GWO), pathfinder algorithm (PFA), \(\varepsilon \) ε -Greedy, and UCB, respectively. This innovative approach has the potential to significantly enhance the accuracy and efficiency of feature selection in medical applications, enabling the development of more reliable and interpretable predictive models while minimizing the use of sensitive patient data.