Scholarlink

<p>Most outlier detection methods identify anomalous observations but provide limited insight into their spatial clustering patterns and reliability. This study introduces a spatial anomaly detection framework that systematically localizes outlier concentrations within interpretable grid regions and quantifies their statistical reliability. The approach used in this study combines the Region Partitioning Framework (RPF) for spatial subdivision with novel Region-Level Anomaly Rate (RAR) metrics that integrate outlier density and deviation severity. Unlike existing methods that produce binary classifications or abstract scores, the proposed RPF–RAR framework provides transparent spatial context through region-based analysis with explicit reliability assessment. Evaluation on biological morphological data (Iris flower measurements) and longitudinal growth data (ChickWeight dataset) demonstrates the framework’s ability to distinguish meaningful spatial patterns from statistical noise, with grid resolution adapting to data structure. The RPF–RAR framework offers a practical, interpretable alternative for applications requiring clear spatial understanding of anomalous behavior.</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

Outlier detection using region partitioning framework and region-level anomaly rate

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Eungi Kim

摘要

Most outlier detection methods identify anomalous observations but provide limited insight into their spatial clustering patterns and reliability. This study introduces a spatial anomaly detection framework that systematically localizes outlier concentrations within interpretable grid regions and quantifies their statistical reliability. The approach used in this study combines the Region Partitioning Framework (RPF) for spatial subdivision with novel Region-Level Anomaly Rate (RAR) metrics that integrate outlier density and deviation severity. Unlike existing methods that produce binary classifications or abstract scores, the proposed RPF–RAR framework provides transparent spatial context through region-based analysis with explicit reliability assessment. Evaluation on biological morphological data (Iris flower measurements) and longitudinal growth data (ChickWeight dataset) demonstrates the framework’s ability to distinguish meaningful spatial patterns from statistical noise, with grid resolution adapting to data structure. The RPF–RAR framework offers a practical, interpretable alternative for applications requiring clear spatial understanding of anomalous behavior.