Scholarlink

<p>To enhance the feature-extraction ability of kernel-based broad learning system (BLS) and thereby improve its overall performance, the paper proposes an Adaboost selective ensemble learning algorithm based on multi-kernel BLS (MKBLS). AS-MKBLS integrates multi-kernel learning at both the feature and decision levels, comprising three main components. First, recognizing the limited feature-mapping capabilities of traditional single-kernel BLS and the challenges posed by manual parameter tuning, a multi-kernel BLS algorithm is designed. By fusing global and local kernel functions to capture diverse feature information, MKBLS achieves composite feature mapping ability, enabling better data representation in high-dimensional feature space, thereby improving model prediction accuracy. Next, to overcome the limitations of feature-level fusion in multi-kernel functions, MKBLS serves as the base model for an ensemble algorithm, designed to fuse the multi-kernel functions at the decision level. Finally, addressing the issue of traditional selective ensemble methods relying on a single measurement metric, a multi-dimensional selective ensemble algorithm based on MKBLS is proposed. Combining performance metrics of accuracy and diversity helps select both accurate and complementary base MKBLS, thereby enhancing the overall performance of the selective ensemble model. Experiments demonstrate the superiority of the proposed method, highlighting its advancements over other kernel-based learners and ensemble methods.</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

Adaboost selective learning based on multi-kernel broad learning system

  • 期刊论文

  • 发表日期：

    2026年5月19日

查看全文
  • Fan Yun,
  • Zhiwen Yu,
  • Kaixiang Yang,
  • C. L. Philip Chen

摘要

To enhance the feature-extraction ability of kernel-based broad learning system (BLS) and thereby improve its overall performance, the paper proposes an Adaboost selective ensemble learning algorithm based on multi-kernel BLS (MKBLS). AS-MKBLS integrates multi-kernel learning at both the feature and decision levels, comprising three main components. First, recognizing the limited feature-mapping capabilities of traditional single-kernel BLS and the challenges posed by manual parameter tuning, a multi-kernel BLS algorithm is designed. By fusing global and local kernel functions to capture diverse feature information, MKBLS achieves composite feature mapping ability, enabling better data representation in high-dimensional feature space, thereby improving model prediction accuracy. Next, to overcome the limitations of feature-level fusion in multi-kernel functions, MKBLS serves as the base model for an ensemble algorithm, designed to fuse the multi-kernel functions at the decision level. Finally, addressing the issue of traditional selective ensemble methods relying on a single measurement metric, a multi-dimensional selective ensemble algorithm based on MKBLS is proposed. Combining performance metrics of accuracy and diversity helps select both accurate and complementary base MKBLS, thereby enhancing the overall performance of the selective ensemble model. Experiments demonstrate the superiority of the proposed method, highlighting its advancements over other kernel-based learners and ensemble methods.