Scholarlink

<p>Biclustering is an effective technique in data mining and pattern recognition. However, traditional biclustering algorithms face two limitations in high-dimensional settings: (1) distance concentration leads to ineffective similarity measures and (2) linear dimensionality reduction disrupts local structural patterns. To address these issues, we propose BCBOF, biclustering based on orthogonal factors. BCBOF constructs orthogonal factors in the vector space of the data, clusters the data using their coordinates in the orthogonal subspace, and derives biclustering results. This approach mitigates data sparsity by reducing dimensionality prior to clustering. We apply BCBOF to stock technical indicators and trend prediction, transforming biclustering results into fuzzy rules and incorporating profit-preserving and stop-loss rules into a fuzzy inference system. Comparative experiments show that BCBOF outperforms existing methods on multiple metrics. Virtual trading experiments on 10 A-share stocks demonstrate that the resulting trading strategies achieve higher returns.</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

Orthogonal factor-based biclustering algorithm (BCBOF) for high-dimensional data and its application in stock trend prediction

  • 期刊论文

  • 发表日期：

    2026年4月24日

查看全文
  • Yan Huang,
  • Da-Qing Zhang

摘要

Biclustering is an effective technique in data mining and pattern recognition. However, traditional biclustering algorithms face two limitations in high-dimensional settings: (1) distance concentration leads to ineffective similarity measures and (2) linear dimensionality reduction disrupts local structural patterns. To address these issues, we propose BCBOF, biclustering based on orthogonal factors. BCBOF constructs orthogonal factors in the vector space of the data, clusters the data using their coordinates in the orthogonal subspace, and derives biclustering results. This approach mitigates data sparsity by reducing dimensionality prior to clustering. We apply BCBOF to stock technical indicators and trend prediction, transforming biclustering results into fuzzy rules and incorporating profit-preserving and stop-loss rules into a fuzzy inference system. Comparative experiments show that BCBOF outperforms existing methods on multiple metrics. Virtual trading experiments on 10 A-share stocks demonstrate that the resulting trading strategies achieve higher returns.