Explainable Evidential Clustering
Victor F. Lopes de Souza, Karima Bakhti, Sofiane Ramdani, Denis Mottet, Abdelhak Imoussaten
TL;DR
This paper introduces a method for explaining evidential clustering results using decision trees, addressing the challenge of interpretability in uncertain data, and validates it on synthetic and real datasets with high satisfaction rates.
Contribution
It develops a novel explainability framework for evidential clustering, including the IEMM algorithm, to generate interpretable explanations considering uncertainty and decision-maker preferences.
Findings
The IEMM algorithm effectively explains evidential clustering results.
Explanations achieved up to 93% satisfaction in validation.
The approach handles uncertainty and partial labeling in explanations.
Abstract
Unsupervised classification is a fundamental machine learning problem. Real-world data often contain imperfections, characterized by uncertainty and imprecision, which are not well handled by traditional methods. Evidential clustering, based on Dempster-Shafer theory, addresses these challenges. This paper explores the underexplored problem of explaining evidential clustering results, which is crucial for high-stakes domains such as healthcare. Our analysis shows that, in the general case, representativity is a necessary and sufficient condition for decision trees to serve as abductive explainers. Building on the concept of representativity, we generalize this idea to accommodate partial labeling through utility functions. These functions enable the representation of "tolerable" mistakes, leading to the definition of evidential mistakeness as explanation cost and the construction of…
Peer Reviews
Decision·Submitted to ICLR 2026
The evidential clustering framework is quite interesting, which I believe already provides more interpretability than hard clustering (since the mass function encodes relevant uncertainty in the clusters). The problem of fitting axis-aligned decision trees provides an additional layer of explainability to the problem. The problem of explainable evidential clustering is also mathematically interesting (at least to those working on explainable clustering) because the criteria from clustering cha
The main drawback of the paper is that the problem is not well-formulated. A few important missing pieces are noted below - The paper proposes IEMM as an approach for explainable evidential clustering without precisely stating the problem (that is, what do we want to minimise while ensuring explainability). One can contrast this with IMM and the corresponding line of work, where the explainable k-means clustering is posed as a problem of achieving low k-means cost while ensuring explainability.
S1. It seems to be the first approach to explain evidential clustering. S2. Several novel notions and definitions are included (e.g. cautious explainer) S3. The proposed IMM extension is well-formulated.
W1. Evidential clustering has not been widely accepted, especially in real applications (compared for example to fuzzy or probabilistic clustering methods). Hence an explainer specialized to that framework has limited reach. W2. If the number of focal elements is large, it seems difficult to interpret the results.
1. The paper tackles the largely unexplored challenge of explainability for evidential clustering, a problem that is of interest in high-stakes and risk-averse domains. 2. It reconstructs IMM’s logic for hard cluster explainers, formalizes representativeness in the evidential regime, and proves that minimizing evidential mistakeness yields maximal representativeness under a stakeholder utility. 3. The use of a stakeholder-specific utility to mediate caution versus specificity is conceptually a
1. The DSClustering paper (Hovhannisyan, 2025) recently proposed a system that also leverages Dempster–Shafer theory to generate interpretable, rule-based cluster descriptions and to communicate uncertainty to end users. Given this development, the authors of the current submission risk slightly overstating the claim that “no one has addressed interpretability in evidential clustering.” While their approach remains distinct, the paper should avoid implying exclusivity in combining Dempster–Shafe
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsSemantic Web and Ontologies · Data Mining Algorithms and Applications