Scholarlink

<p>Support vector machine (SVM), classification, and clustering are interconnected concepts within the fields of machine learning and data analysis, often utilizing kernel functions to effectively manage nonlinear data. This paper introduces a novel approach to mathematical modeling of dynamic data sets and considers the SVM kernel, based on hybrid similarities. We propose “angle–distance-based kernels” that simultaneously measure the similarity between data subsets based on their relative angles and modified distances with respect to a set of fixed reference vectors. This approach enables SVMs to capture clusters of data characterized by shared directional trends rather than relying solely on one-dimensional estimators. We discuss the effectiveness of combined-based kernels on real-world data sets and explain how they can enhance SVM performance in the situation where multiple relationships are crucial. One notable strength of this design is its excellent adaptability for real-time data analysis with high accuracy.</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

A combined kernel function for dynamic support vector machines: exploiting hybrid similarity in data clustering

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Hamid Reza Goudarzi,
  • Zahra Farzinrad

摘要

Support vector machine (SVM), classification, and clustering are interconnected concepts within the fields of machine learning and data analysis, often utilizing kernel functions to effectively manage nonlinear data. This paper introduces a novel approach to mathematical modeling of dynamic data sets and considers the SVM kernel, based on hybrid similarities. We propose “angle–distance-based kernels” that simultaneously measure the similarity between data subsets based on their relative angles and modified distances with respect to a set of fixed reference vectors. This approach enables SVMs to capture clusters of data characterized by shared directional trends rather than relying solely on one-dimensional estimators. We discuss the effectiveness of combined-based kernels on real-world data sets and explain how they can enhance SVM performance in the situation where multiple relationships are crucial. One notable strength of this design is its excellent adaptability for real-time data analysis with high accuracy.