Scholarlink

<p>Support Vector Clustering (SVC) is a powerful unsupervised learning method for detecting arbitrarily shaped clusters, making it highly relevant for complex pattern analysis tasks. However, its widespread use has been constrained by the high computational cost associated with its labeling phase. This paper presents three novel methods that enhance the scalability and efficiency of SVC, making it more suitable for large-scale pattern recognition applications. We first propose the Inheritance Support Vector Graph (InSVG), an optimized graph construction strategy that reduces redundant connectivity checks through transitive inference. We then introduce two intelligent labeling heuristics: Most Similar Test First (MSTF), which prioritizes similarity-based evaluations, and Most Similar Label First (MSLF), a fast approximation that eliminates connection tests by leveraging kernel-based proximity. These methods are evaluated on 15 diverse datasets, demonstrating labeling time reductions of up to 97% while preserving clustering quality, as measured by standard internal and external metrics. The proposed framework contributes to the field of pattern analysis by enabling the practical deployment of SVC in real-time or large-scale scenarios involving high-dimensional data.</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

Enhancing support vector clustering labeling efficiency with scalable heuristics

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Abou Bakr Seddik Drid,
  • Abdelhamid Djeffal,
  • Abdelmalik Taleb-Ahmed

摘要

Support Vector Clustering (SVC) is a powerful unsupervised learning method for detecting arbitrarily shaped clusters, making it highly relevant for complex pattern analysis tasks. However, its widespread use has been constrained by the high computational cost associated with its labeling phase. This paper presents three novel methods that enhance the scalability and efficiency of SVC, making it more suitable for large-scale pattern recognition applications. We first propose the Inheritance Support Vector Graph (InSVG), an optimized graph construction strategy that reduces redundant connectivity checks through transitive inference. We then introduce two intelligent labeling heuristics: Most Similar Test First (MSTF), which prioritizes similarity-based evaluations, and Most Similar Label First (MSLF), a fast approximation that eliminates connection tests by leveraging kernel-based proximity. These methods are evaluated on 15 diverse datasets, demonstrating labeling time reductions of up to 97% while preserving clustering quality, as measured by standard internal and external metrics. The proposed framework contributes to the field of pattern analysis by enabling the practical deployment of SVC in real-time or large-scale scenarios involving high-dimensional data.