Scholarlink

<p>Classification methods for mixed-attribute data typically adopt a two-stage strategy, which first either converts such data into a uniform type or constructs a distance metric, and then applies the result to downstream classification models. However, these methods overlook the inherent ordering of ordinal categorical values during the first stage data processing or distance metric, and their two stages are handled separately without interaction, exhibiting insufficient compatibility. To address these issues, this paper extends the RBF-ELM (Radial Basis Function - Extreme Learning Machine) classifier to directly (without stage-wise) handle mixed-attribute data containing ordinal attributes, and proposes a Meta-Learning-driven RBF-ELM classifier (ML-RBF-ELM for short). Specifically, by mining the associations between class labels and categorical attributes (including nominal and ordinal types), we design a novel conditional entropy-based distance metric to compute the distances between input samples and kernel centers in RBF-ELM, thereby extending RBF-ELM into a classification model capable of directly processing mixed-attribute data. Furthermore, to effectively select RBF kernel centers, we employ a meta-learning method to adaptively determine the optimal hyperparameter <i>k</i> in <i>k</i>-prototypes clustering, and subsequently select the optimal RBF kernel centers. Experimental results on 30 mixed-attribute datasets demonstrate that our ML-RBF-ELM achieves a favorable performance trade-off between classification accuracy and computational efficiency compared with nine state-of-the-art competing 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

A meta-learning RBF-ELM network with a novel conditional entropy distance metric for mixed-attribute data classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月11日

查看全文
  • Qiude Li,
  • Liezhen Wang,
  • Shengfen Ji,
  • Yang Yu,
  • Sigui Hu,
  • Qingyu Xiong,
  • Zuquan Hu,
  • Zhu Zeng

摘要

Classification methods for mixed-attribute data typically adopt a two-stage strategy, which first either converts such data into a uniform type or constructs a distance metric, and then applies the result to downstream classification models. However, these methods overlook the inherent ordering of ordinal categorical values during the first stage data processing or distance metric, and their two stages are handled separately without interaction, exhibiting insufficient compatibility. To address these issues, this paper extends the RBF-ELM (Radial Basis Function - Extreme Learning Machine) classifier to directly (without stage-wise) handle mixed-attribute data containing ordinal attributes, and proposes a Meta-Learning-driven RBF-ELM classifier (ML-RBF-ELM for short). Specifically, by mining the associations between class labels and categorical attributes (including nominal and ordinal types), we design a novel conditional entropy-based distance metric to compute the distances between input samples and kernel centers in RBF-ELM, thereby extending RBF-ELM into a classification model capable of directly processing mixed-attribute data. Furthermore, to effectively select RBF kernel centers, we employ a meta-learning method to adaptively determine the optimal hyperparameter k in k-prototypes clustering, and subsequently select the optimal RBF kernel centers. Experimental results on 30 mixed-attribute datasets demonstrate that our ML-RBF-ELM achieves a favorable performance trade-off between classification accuracy and computational efficiency compared with nine state-of-the-art competing methods.