Scholarlink

<p>Takagi–Sugeno–Kang (TSK) fuzzy classifier, due to its simplicity in computation, ease of implementation, and inherent interpretability, has been studied for handling various uncertainties. However, most existing variants of TSK classifier face the following challenges: (1) high coupling between features, (2) difficulties in expressing the importance of features in fuzzy rule setting, and (3) neglect of the transmission of important rules in the decision-making process. In this study, a novel weighted-based hierarchical TSK fuzzy classifier, WB-D-TSK-FC, is constructed to address the above challenges. A feature scoring mechanism and a short-rule antecedent parameters optimization strategy with enlarged weight of fuzzy membership expectation value are proposed to weaken the coupling relationship between features and reduce computational complexity. An error-active intervention is proposed to constrain the training errors of each sub-classifier to improve the classifier’s generalization ability. Furthermore, an important rule transfer fusion mechanism is designed to fully exploit the decision-making ability of important rules and their crucial roles in information transmission. Finally, an input space optimization is proposed, which emphasizes continuous historical judgment information without deviating from the original training space, aiming to maintain high interpretability. Experimental results show that WB-D-TSK-FC indeed exhibits strong classification advantages compared to several selected classifiers.</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

Interpretable TSK Fuzzy Classification with Preserved Approximate Physical Properties for Source Space Features

  • 期刊论文

  • 发表日期：

    2025年6月29日

查看全文
  • Jinghao Chen,
  • Jia Zhai,
  • Yuanqing Yang,
  • Guanghao Song,
  • Ta Zhou

摘要

Takagi–Sugeno–Kang (TSK) fuzzy classifier, due to its simplicity in computation, ease of implementation, and inherent interpretability, has been studied for handling various uncertainties. However, most existing variants of TSK classifier face the following challenges: (1) high coupling between features, (2) difficulties in expressing the importance of features in fuzzy rule setting, and (3) neglect of the transmission of important rules in the decision-making process. In this study, a novel weighted-based hierarchical TSK fuzzy classifier, WB-D-TSK-FC, is constructed to address the above challenges. A feature scoring mechanism and a short-rule antecedent parameters optimization strategy with enlarged weight of fuzzy membership expectation value are proposed to weaken the coupling relationship between features and reduce computational complexity. An error-active intervention is proposed to constrain the training errors of each sub-classifier to improve the classifier’s generalization ability. Furthermore, an important rule transfer fusion mechanism is designed to fully exploit the decision-making ability of important rules and their crucial roles in information transmission. Finally, an input space optimization is proposed, which emphasizes continuous historical judgment information without deviating from the original training space, aiming to maintain high interpretability. Experimental results show that WB-D-TSK-FC indeed exhibits strong classification advantages compared to several selected classifiers.