Scholarlink

<p>High-dimensional classification, where the number of features far exceeds the sample size, requires effective and stable feature selection. Penalized logistic regression is a popular choice, but it often produces unstable results that are sensitive to training data splits and tuning parameters. We propose a two-stage method, <i>Frequency-Based Ranking and Incremental Feature Selection</i>, to improve selection stability and classification performance. First, features are ranked by their selection frequencies over <i>N</i> repeated penalized logistic regressions. Second, a chosen classifier is applied using an incremental inclusion of ranked features, with performance evaluated across repeated splits. Simulation studies and real data analyses are conducted to demonstrate the finite-sample performance of the proposed method.</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

Stability-ranked feature selection for classification in high-dimensional data: combining regularization and machine learning algorithms

  • 期刊论文

  • 发表日期：

    2026年4月21日

查看全文
  • Mohammad Kazemi,
  • Amirhossein Khadivi Noghredeh

摘要

High-dimensional classification, where the number of features far exceeds the sample size, requires effective and stable feature selection. Penalized logistic regression is a popular choice, but it often produces unstable results that are sensitive to training data splits and tuning parameters. We propose a two-stage method, Frequency-Based Ranking and Incremental Feature Selection, to improve selection stability and classification performance. First, features are ranked by their selection frequencies over N repeated penalized logistic regressions. Second, a chosen classifier is applied using an incremental inclusion of ranked features, with performance evaluated across repeated splits. Simulation studies and real data analyses are conducted to demonstrate the finite-sample performance of the proposed method.