Scholarlink

The field of Explainable AI (XAI) has largely focused on interpretability in classification and regression, while cluster analysis, a vital unsupervised machine learning technique, has been relatively neglected. As an exploratory technique, the interpretability and explainability of modern clustering models present significant challenges. This lack of transparency impedes trust and comprehension of the models’ decision-making processes. In this paper, we address this gap by enhancing the interpretability of clustering results using a hierarchical structure representation. Our approach employs oblique decision trees, supported by SHAP (SHapley Additive exPlanations) values to analyze influential features and identify separation hyperplanes for constructing the trees. Unlike traditional axis-aligned trees, oblique decision trees provide a more accurate interpretation of high-dimensional data, maintaining a clear and interpretable structure. This method not only improves transparency but also fosters trust in hybrid decision-making systems, offering reliable and comprehensible explanations for clustering outcomes.

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

Trustworthy Clustering: An Interpretable Clustering Framework with Hierarchical Oblique Decision Boundaries

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lige Gan

摘要

The field of Explainable AI (XAI) has largely focused on interpretability in classification and regression, while cluster analysis, a vital unsupervised machine learning technique, has been relatively neglected. As an exploratory technique, the interpretability and explainability of modern clustering models present significant challenges. This lack of transparency impedes trust and comprehension of the models’ decision-making processes. In this paper, we address this gap by enhancing the interpretability of clustering results using a hierarchical structure representation. Our approach employs oblique decision trees, supported by SHAP (SHapley Additive exPlanations) values to analyze influential features and identify separation hyperplanes for constructing the trees. Unlike traditional axis-aligned trees, oblique decision trees provide a more accurate interpretation of high-dimensional data, maintaining a clear and interpretable structure. This method not only improves transparency but also fosters trust in hybrid decision-making systems, offering reliable and comprehensible explanations for clustering outcomes.