Scholarlink

<p>Attribute reductions facilitate data analyses and knowledge discovery, and they rely on uncertainty measures. The current dependency degree (DD) and condition entropy (CE) effectively motivate attribute reductions and corresponding heuristic algorithms (called DD-AR and CE-AR). However, algebraic DD and informational CE emphasize only a single view, and their algebra-information fusion implies advancement space for uncertainty measurements and attribute reductions. For measurement reinforcement, the roughness/accuracy degree (RD/AD) is further introduced to constitute three-factor measures DD, CE, RD/AD with two levels, the direct classification-level fusion and preferential class-level fusion produce two fusion measures (called DFM and PFM) with hierarchical integrations, so DFM and PFM motivate attribute reductions and corresponding heuristic algorithms (called DFM-AR and PFM-AR) with learning improvements. At first, hierarchical sizes of RDs/ADs are revealed by max-min and mean statistics of class-level values. Then, DFM directly emerges by fused multiplication of classification-level three-factor measures, while PFM hierarchically emerges by fused multiplication of class-level three-factor measures and by subsequent integrated summation; DFM and PFM acquire essential expressions, size relationships, granulation monotonicity, and construction algorithms. Furthermore, DFM and PFM generate attribute reductions, and DFM-AR and PFM-AR are designed by attribute significance to improve DD-AR and CE-AR. Finally, relevant fusion measures and reduction algorithms are validated through table examples and data experiments, so the two new algorithms generally outperform current baselines DD-AR, CE-AR and other latest algorithms (such as AR-JSI, UDI-IG) to achieve better classification performances.</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

Improved attribute reductions based on combined measures from dependency-roughness-entropy fusion and classification-class-level integration

  • 期刊论文

  • 发表日期：

    2025年6月16日

查看全文
  • Yixiao Yuan,
  • Xianyong Zhang,
  • Qian Wang,
  • Zhiwen Mo

摘要

Attribute reductions facilitate data analyses and knowledge discovery, and they rely on uncertainty measures. The current dependency degree (DD) and condition entropy (CE) effectively motivate attribute reductions and corresponding heuristic algorithms (called DD-AR and CE-AR). However, algebraic DD and informational CE emphasize only a single view, and their algebra-information fusion implies advancement space for uncertainty measurements and attribute reductions. For measurement reinforcement, the roughness/accuracy degree (RD/AD) is further introduced to constitute three-factor measures DD, CE, RD/AD with two levels, the direct classification-level fusion and preferential class-level fusion produce two fusion measures (called DFM and PFM) with hierarchical integrations, so DFM and PFM motivate attribute reductions and corresponding heuristic algorithms (called DFM-AR and PFM-AR) with learning improvements. At first, hierarchical sizes of RDs/ADs are revealed by max-min and mean statistics of class-level values. Then, DFM directly emerges by fused multiplication of classification-level three-factor measures, while PFM hierarchically emerges by fused multiplication of class-level three-factor measures and by subsequent integrated summation; DFM and PFM acquire essential expressions, size relationships, granulation monotonicity, and construction algorithms. Furthermore, DFM and PFM generate attribute reductions, and DFM-AR and PFM-AR are designed by attribute significance to improve DD-AR and CE-AR. Finally, relevant fusion measures and reduction algorithms are validated through table examples and data experiments, so the two new algorithms generally outperform current baselines DD-AR, CE-AR and other latest algorithms (such as AR-JSI, UDI-IG) to achieve better classification performances.