Scholarlink

<p>Attribute reduction or feature selection plays an important role in improving performance and reducing complexity of models. Rough sets have been widely applied to such task because of a natural advantage of modeling abilities of the uncertain and imprecise data. But, increasingly complex data bring three challenges to them: limited labeled and mixed-attribute property of big data, computational inefficiency and overfitting in attribute reduction. To address these challenges, this paper proposed a generalized local neighborhood rough set model (LNRS). With the proposed model, we develop an efficient attribute reduction algorithm for mixed data with limited labels and its efficiency is demonstrated on nine UCI data sets. The theoretical analysis and experimental results show that the proposed LNRS and its corresponding algorithms significantly outperform its original counterpart, suggesting that the local neighborhood model based methods are more flexible to deal with mixed data with limited labels.</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

Mixed attribute reduction in limited labeled data: a local neighborhood rough set approach

  • 期刊论文

  • 发表日期：

    2025年8月17日

查看全文
  • Qian Guo,
  • Xinyan Liang,
  • Zhihua Cui,
  • Junyan Wang,
  • Zhaoyi Wu

摘要

Attribute reduction or feature selection plays an important role in improving performance and reducing complexity of models. Rough sets have been widely applied to such task because of a natural advantage of modeling abilities of the uncertain and imprecise data. But, increasingly complex data bring three challenges to them: limited labeled and mixed-attribute property of big data, computational inefficiency and overfitting in attribute reduction. To address these challenges, this paper proposed a generalized local neighborhood rough set model (LNRS). With the proposed model, we develop an efficient attribute reduction algorithm for mixed data with limited labels and its efficiency is demonstrated on nine UCI data sets. The theoretical analysis and experimental results show that the proposed LNRS and its corresponding algorithms significantly outperform its original counterpart, suggesting that the local neighborhood model based methods are more flexible to deal with mixed data with limited labels.