Pith. sign in

REVIEW 3 major objections 5 minor 72 references

Explaining Black-Box Clustering Pipelines With Cluster-Explorer

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that Cluster-Explorer explains any black-box clustering pipeline by mining concise, Pareto-optimal conjunctions of predicates via generalized frequent-itemset mining, and that these explanations beat XAI baselines in…

desk verdict A solid systems paper with a clean gFIM reduction, but the headline quality claim is partly circular and the abstract overstates the runtime advantage; it deserves peer review with revisions. read the letter →

arxiv 2412.20446 v1 pith:Q7FN5DLY submitted 2024-12-29 cs.DB

Sariel Ofek , Amit Somech This is my paper · ORCID
classification cs.DB
keywords clusterexplanationblack-boxclusteringgeneralizedfrequentitemsetminingexplainableAIrule-basedexplanationsParetooptimalityattributeselectionpost-hocexplainability
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that the results of any clustering pipeline—whatever preprocessing and algorithm was used—can be explained automatically by short, human-readable rules. It proposes Cluster-Explorer, which mines conjunctions of predicates such as 'Age between 16 and 35 and Education-num between 4 and 13,' choosing rules that cover most of a cluster's points, rarely touch other clusters, and stay short. The authors argue this works for any black-box pipeline because the tool never needs an auxiliary supervised model or an interpretable-by-design clustering algorithm. On 98 clustering results spanning five algorithms and 19 datasets, they report higher explanation quality than four XAI baselines, with an average runtime of 55.8 seconds and a 14.4x speedup from attribute selection.

What carries the argument

The load-bearing object is the reduction to generalized frequent itemset mining (gFIM) over augmented transactions. gFIM extends classic frequent itemset mining by letting items be organized in a taxonomy, so a mined itemset can contain either a specific value or a more general category; here the category hierarchy is an interval taxonomy built from multiple binning methods, and items are predicates. The reduction makes coverage equal to itemset frequency, lets the taxonomy eliminate overlapping interval predicates, and lets a high support threshold and small maximal itemset size keep the exponential mining cost manageable; a Gini-importance-based attribute selection step further prunes items.

What would settle it

Build a synthetic dataset where one cluster is deliberately two well-separated modes (for example, ages 18-25 and 55-65 with no other distinguishing attributes) and run Cluster-Explorer. Any single conjunction of interval predicates can cover at most one mode, so the best rule's coverage will be near 50% and the QSE will fall far below the reported 0.84 average; this directly tests the central quality claim.

Watch

Extended reading notes

Core claim

At the paper's center is a reformulation: explaining a cluster is the same as mining frequent generalized itemsets. Each data point becomes an augmented transaction whose items are predicates—numeric intervals from five binning methods, raw values, and negations of categorical values—organized in an interval taxonomy. Running generalized frequent itemset mining separately on each cluster's transactions yields candidate explanations whose frequency is exactly coverage; candidates are then filtered by separation error and reduced to Pareto-optimal explanations trading off coverage, separation error, and conciseness. The paper's empirical claim is that on its 98-instance benchmark this pipeline produces explanations with an average QSE of 0.84 versus 0.72 for the best baseline, that quality degrades far less as rows, columns, or clusters grow, and that a decision-tree-based attribute selection step cuts runtime by 14.4x with negligible quality loss.

Load-bearing premise

The system assumes every cluster worth explaining can be captured by one short conjunction of numeric intervals and categorical negations; a cluster that splits into two or more disjoint profiles has no concise explanation in this language, and the reported quality does not cover that case.

Editorial extensions

If this is right

  • Any clustering pipeline can be explained post hoc without retraining: only the original data and cluster labels are needed.
  • Explanations come out as short conjunctions of range and inequality predicates, so non-experts can read and sanity-check them directly.
  • On the 98-result benchmark the reported average QSE beats all four baselines and stays nearly flat as dataset size, width, and cluster count grow.
  • The attribute-selection optimization gives an average 14.4x runtime reduction, reaching up to 26.9x on large datasets, with under 0.1 QSE loss.
  • Because quality holds across five different clustering algorithms, the tool does not force users to switch to an interpretable clustering algorithm.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the explanation language is limited to conjunctions of intervals and negations, a cluster that is really two disjoint profiles will need several rules or a disjunction; adding disjunctions would be a direct extension of the gFIM encoding.
  • The equal weighting of coverage, separation error, and conciseness in QSE is a modeling choice; re-ranking the same Pareto-optimal front under different weights would let users emphasize auditability versus discovery without re-running the miner.
  • The same attribute-selection trick should transfer to other high-coverage, short-pattern mining tasks such as subgroup discovery, where Gini-based pruning could cut the itemset search space similarly.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. Cluster-Explorer is a post-hoc explainability tool for black-box clustering pipelines. An explanation is a conjunction of predicates over numeric interval bins and categorical negations, and the paper defines explanation quality in terms of coverage, separation error, and conciseness. The core algorithmic contribution is a reduction to generalized frequent-itemset mining (gFIM): for each cluster, Cluster-Explorer mines frequent itemsets with minimum support equal to the coverage threshold and maximum size equal to the inverse conciseness threshold, converts itemsets to predicate conjunctions, filters by separation error, and then computes Pareto-optimal explanations via the skyline operator. An attribute-selection step based on decision-tree Gini importance is added for scalability. The paper evaluates the method on 98 clustering instances derived from 19 UCI datasets and 16 pipelines, comparing against SHAP, Anchors, SkopeRules, and decision-tree paths on a composite QSE score and on runtime, and it reports a small user study.

Significance. If the quality claims hold, Cluster-Explorer would be a practically useful, algorithm-agnostic tool for obtaining rule-like cluster explanations at scale, with experiments covering datasets up to 1.7M rows. The gFIM reduction is clean and the implementation, benchmark, and code are publicly available, which is a tangible contribution to the community. However, the evaluation of explanation quality is substantially weakened by the fact that the automatic metric QSE is the same objective that Cluster-Explorer optimizes, and by the small and partial user study. The runtime claim in the abstract is also overstated relative to the paper's own results. The algorithmic machinery and the benchmark are solid, but the central 'superiority' claim needs additional, non-circular evidence.

major comments (3)
  1. [Section 5.1 and Section 4.3] The automatic quality evaluation uses QSE = (Coverage + (1 - SeparationErr) + Conciseness) / 3, and Algorithm 2 generates candidate explanations by enforcing Coverage >= theta_cov and Conciseness >= theta_con, filtering SeparationErr <= theta_sep, and then selecting the skyline over exactly these three quantities. The reported QSE gap (e.g., 0.84 for Cluster-Explorer vs. 0.72 for SkopeRules in Section 5.2.1) therefore largely measures how closely each method matches Cluster-Explorer's own objective, not an independent notion of explanation quality. The claim in Section 5 that 'the explanations generated by Cluster-Explorer are superior to those of the baselines in terms of ... quality' should be supported by an evaluation criterion that is not the optimized objective, such as a larger user study, a downstream task, or a comparison at matched conciseness levels.
  2. [Abstract and Section 5.2.4] The abstract claims superiority over XAI baselines in execution times, but Section 5.2.4 reports the Decision Tree baseline at 6.3 seconds versus 55.8 seconds for Cluster-Explorer. The text later qualifies that Decision Tree is not quality-competitive and that Cluster-Explorer is 12.6x faster than the closest quality-competitive baseline, but the unqualified statement in the abstract is inaccurate. The runtime claim should be restated to refer to quality-competitive baselines, or the comparison should include a quality-adjusted runtime.
  3. [Section 5.2.2] The user study is the only non-circular evidence for quality superiority, but it is small and partial: 12 participants, three datasets, and 36 evaluations by the paper's own count, with SHAP and Anchors excluded. The description is also internally inconsistent: 'each participant evaluated three cluster explanations per dataset' over three datasets with 12 participants would yield 108 evaluations, not 36. As reported, this study cannot by itself support the strong claim that Cluster-Explorer produces superior explanations; it should be presented as preliminary evidence or the claim should be softened.
minor comments (5)
  1. [Section 3.3, Definition 3.3] Condition (2) is missing a negation symbol in the typeset formula; it should read 'there does not exist E_c such that ...' to express Pareto optimality correctly.
  2. [Section 6] The paper acknowledges that disjunctions are future work, but the abstract and Section 5 should state that the claimed superiority is within the restricted explanation language of conjunctions of interval and negation predicates; clusters with disjoint profiles cannot be expressed in this language.
  3. [Example 1.2] The text says 'Figure 1 depicts two example cluster explanations', but Figure 1 shows clustering results and Figure 2 shows the explanations; the cross-reference is incorrect.
  4. [Section 5.2.2] The count of user-study evaluations is ambiguous and should be clarified, along with whether each participant rated all three datasets or a subset.
  5. [General editorial] There are several typos, including 'unclear clear' in Section 1, 'in order too obtain' in Section 4.3, and formatting issues in Algorithm 1 and Figure 3 captions.

Circularity Check

1 steps flagged · score 6.0 of 10

Automatic quality superiority is measured by QSE, a simple average of exactly the three criteria Cluster-Explorer enforces and Pareto-optimizes in Algorithm 2, so the headline quality gap over baselines is partly an artifact of evaluating the method on its own objective; only a small, partial user study provides independent support.

  1. self definitional [Section 5.1 (Evaluating Metrics), compared with Definition 3.3 and Algorithm 2 (Section 4.3)]
    "Based on the clustering explanations quality criteria described in Section 3.2, we use a unified Quality Score for an Explanation (QSE) which balances the coverage, separation error, and conciseness. QSE is defined for a single cluster explanation by: QSE(𝐸𝑐) = Coverage(𝐸𝑐)+ (1− SeparationErr(𝐸𝑐))+ Conciseness(𝐸𝑐) / 3"

    The three components of QSE are precisely the three objectives that Cluster-Explorer is built to optimize. Algorithm 2 runs gFIM with minsup = θ_cov and maxsize = 1/θ_con, filters explanations by SeparationErr ≤ θ_sep, and then selects via SKYLINE(Coverage, SepError, Conciseness). Definition 3.3 defines the desired explanation set using exactly the same coverage, separation-error, and conciseness Pareto criteria. Thus every explanation Cluster-Explorer returns is constrained to score highly on QSE by construction. The reported 0.84 vs. 0.72 advantage over SkopeRules therefore measures, to a substantial degree, how closely each baseline happens to match Cluster-Explorer's own scoring function rather than an externally grounded notion of semantic quality.

full rationale

The central automatic evaluation is partially circular. Section 5.1 defines QSE as the average of coverage, (1 - separation error), and conciseness, and Section 5.2.1 uses this QSE to claim superiority. But Section 4.3 / Algorithm 2 generates Cluster-Explorer's explanations under exactly those criteria: gFIM enforces coverage ≥ θ_cov and itemset size ≤ 1/θ_con, the separation-error filter enforces SeparationErr ≤ θ_sep, and the skyline returns Pareto-optimal explanations with respect to coverage, separation error, and conciseness. So Cluster-Explorer is optimized to maximize the same quantity on which it is evaluated; the quality gap partly reduces to objective alignment. The paper does offer an independent user study, and its results correlate with QSE, but the study is too small and partial to fully carry the universal superiority claim. There is no load-bearing self-citation: references [57] and [58] point to the authors' code and demo but are not used to justify the core derivation, and the gFIM reduction is grounded in standard external literature. Separately, the abstract's claim of superior execution times is contradicted by Section 5.2.4, where Decision Tree is roughly 9x faster than Cluster-Explorer (6.3s vs. 55.8s); this is an overclaim rather than a circularity. Overall, the quality result is partially reduced to the method's own objective, so the circularity score is 6.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard gFIM correctness, the interval-taxonomy containment model, user-chosen quality thresholds, and the decision-tree importance heuristic. No new physical or mathematical entities are postulated. The main freedom is in the threshold choices and the binning method set, which shape the entire predicate space.

free parameters (5)
  • coverage threshold theta_cov = 0.8
    Chosen by the user, used as the gFIM minimum support; directly controls which rules survive.
  • separation error threshold theta_sep = 0.3
    Chosen by the user, filters candidate explanations after gFIM mining.
  • conciseness threshold theta_con = 0.2
    Chosen by the user, sets max itemset size to 1 / theta_con = 5 predicates.
  • attribute selection scaling p = 1
    Controls the number of selected attributes via n_attr = floor((1 / theta_con) * p).
  • binning method set
    Five binning strategies (equal-height, equal-width, 1-D clustering, tree-based, optimal) are chosen by hand and define the set of available numeric predicates.
assumptions (5)
  • standard math Generalized frequent itemset mining algorithms correctly return all itemsets with support above the threshold and size below the maximum.
    The method relies on the correctness of the cited gFIM algorithm family, e.g., Srikant and Agrawal 1997, in Section 4.3.
  • standard math The interval taxonomy partial order correctly represents containment, so the support of an interval itemset equals the coverage of the corresponding range predicate.
    Section 4.2 constructs the taxonomy and connects each value to all containing intervals; this is the bridge between gFIM support and explanation coverage.
  • domain assumption Coverage, separation error, and conciseness as defined in Section 3.2 are appropriate operationalizations of explanation quality.
    The entire evaluation and problem definition depend on these three measures being what users want from an explanation.
  • domain assumption Gini importance from a binary decision tree predicts which attributes are relevant for explaining clusters.
    Section 4.4 uses mean Gini impurity across per-cluster decision trees to select the top attributes for gFIM.
  • domain assumption Filtering clustering results by silhouette coefficient >= 0.1 removes only near-random results and leaves a representative benchmark.
    Section 5.1 applies this filter to all pipelines; if the filter biases the benchmark, the claimed superiority may not generalize.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Explaining Black-Box Clustering Pipelines With Cluster-Explorer." pith.science (2026). https://pith.science/paper/Q7FN5DLY

@misc{pith2026241220446,
  author       = {Pith},
  title        = {Pith review of: Explaining Black-Box Clustering Pipelines With Cluster-Explorer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q7FN5DLY}},
  note         = {Machine review of arXiv:2412.20446}
}
read the original abstract

Explaining the results of clustering pipelines by unraveling the characteristics of each cluster is a challenging task, often addressed manually through visualizations and queries. Existing solutions from the domain of Explainable Artificial Intelligence (XAI) are largely ineffective for cluster explanations, and interpretable-by-design clustering algorithms may be unsuitable when the clustering algorithm does not fit the data properties. To bridge this gap, we introduce Cluster-Explorer, a novel explainability tool for black-box clustering pipelines. Our approach formulates the explanation of clusters as the identification of concise conjunctions of predicates that maximize the coverage of the cluster's data points while minimizing separation from other clusters. We achieve this by reducing the problem to generalized frequent-itemsets mining (gFIM), where items correspond to explanation predicates, and itemset frequency indicates coverage. To enhance efficiency, we leverage inherent problem properties and implement attribute selection to further reduce computational costs. Experimental evaluations on a benchmark collection of 98 clustering results, as well as a user study, demonstrate the superiority of Cluster-Explorer in both explanation quality and execution times compared to XAI baselines.

Figures

Figures reproduced from arXiv: 2412.20446 by the authors.

Figure 1
Figure 1. She sees that the data points are fairly segmented into three [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Example cluster explanations generated by [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Example augmented transactions itemset size, as effective explanations should be concise [41]. The candidate explanations generated by the gFIM algorithm are further filtered to remove those with high separation error. We then select only the Pareto optimal explanations using the skyline operator [6]. Finally, in Section 4.4, we describe a simple yet highly effective optimization technique that reduces running times… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Combined figures illustrating QSE and metrics analysis. lower than Cluster-Explorer, which attains the highest average QSE of 0.84. In Figure 4b we further report the average maximal score ob￾tain for each metric individually. Cluster-Explorer achieves the highest scor…
Figure 5
Figure 5. Figure 5: User Study Results SkopeRules, and Decision-Tree), we additionally described the ex￾planation coverage and separation error rates (phrased as in [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Average QSE scores as a factor of the number of rows, columns, and clusters [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: QSE vs. Num. of Clusters(Simulated) rows, columns, and clusters increase, respectively, reflecting its limitations for large, complex datasets [4, 14]. Other baselines also degrade with data complexity: SkopeRules drops by 14.1%, 11.4%, and 27.4%, Anchors by 35.9%, 51.…
Figure 8
Figure 8. Figure 8: Explanations generation times (for all clusters) as a factor of the number of rows, columns, and clusters. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: The effect of 𝑝 on the QSE as a factor of the number of rows, columns, and clusters (a) Runtime vs. Number of Rows (b) Runtime vs. Number of Columns (c) Runtime vs. Number of Clusters [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: The effect of 𝑝 on running times as a factor of the number of rows, columns, and clusters time from 13.42 minutes to just 55.8 seconds, with minimal impact on QSE scores (<0.1 difference). 6 CONCLUSION Cluster-Explorer is a novel framework for post-hoc, rule-based exp…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 49 canonical work pages

  1. [1]

    Adult Income Dataset (UCI). 2024. https://archive.ics.uci.edu/ml/datasets/Adult/

    work page 2024

  2. [2]

    Rakesh Agrawal, Ramakrishnan Srikant, et al. 1994. Fast algorithms for mining association rules. In Proc. 20th int. conf. very large data bases, VLDB , Vol. 1215. Citeseer, 487–499

    work page 1994

  3. [3]

    Suad A Alasadi and Wesam S Bhaya. 2017. Review of data preprocessing tech- niques in data mining. Journal of Engineering and Applied Sciences 12, 16 (2017), 4102–4107

    work page 2017

  4. [4]

    Alejandro Barredo Arrieta, Natalia Díaz-Rodríguez, Javier Del Ser, Adrien Ben- netot, Siham Tabik, Alberto Barbado, Salvador García, Sergio Gil-López, Daniel Molina, Richard Benjamins, et al. 2020. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. In- formation fusion 58 (2020), 82–115

    2020

  5. [5]

    Szymon Bobek, Michal Kuk, Maciej Szelążek, and Grzegorz J Nalepa. 2022. En- hancing cluster analysis with explainable AI and multidimensional cluster pro- totypes. IEEE Access 10 (2022), 101556–101574

    work page 2022

  6. [6]

    Stephan Borzsony, Donald Kossmann, and Konrad Stocker. 2001. The skyline operator. In Proceedings 17th international conference on data engineering . IEEE, 421–430

    work page 2001

  7. [7]

    Breiman, J

    L. Breiman, J. Friedman, C.J. Stone, and R.A. Olshen. 1984. Classification and Regression Trees. Taylor & Francis

    work page 1984

  8. [8]

    Marco Cavallo and Çağatay Demiralp. 2018. Clustrophile 2: Guided visual clustering analysis. IEEE transactions on visualization and computer graphics 25, 1 (2018), 267–276

    work page 2018

Show all 72 references
  1. [9]

    Yair Censor. 1977. Pareto optimality in multiobjective problems. Applied Mathe- matics and Optimization 4, 1 (1977), 41–59

    1977

  2. [10]

    Angelos Chatzimparmpas, Rafael M Martins, and Andreas Kerren. 2020. t-visne: Interactive assessment and interpretation of t-sne projections. IEEE transactions on visualization and computer graphics 26, 8 (2020), 2696–2714

    2020

  3. [11]

    Tianqi Chen and Carlos Guestrin. 2016. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining . 785–794

    2016

  4. [12]

    William W Cohen and Yoram Singer. 1999. A simple, fast, and effective rule learner. AAAI/IAAI 99, 335-342 (1999), 3

    1999

  5. [13]

    Brian A Davey and Hilary A Priestley. 2002. Introduction to lattices and order . Cambridge university press

    2002

  6. [14]

    Finale Doshi-Velez and Been Kim. 2017. Towards a rigorous science of inter- pretable machine learning. arXiv preprint arXiv:1702.08608 (2017)

    2017 arXiv

  7. [15]

    Mengnan Du, Ninghao Liu, and Xia Hu. 2019. Techniques for interpretable machine learning. Commun. ACM 63, 1 (2019), 68–77

    2019

  8. [16]

    George H Dunteman. 1989. Principal components analysis. Number 69. Sage

    1989

  9. [17]

    Charles A Ellis, Mohammad SE Sendi, Eloy Geenjaar, Sergey M Plis, Robyn L Miller, and Vince D Calhoun. 2021. Algorithm-Agnostic Explainability for Unsu- pervised Clustering. arXiv preprint arXiv:2105.08053 (2021)

    2021 arXiv

  10. [18]

    Martin Ester, Hans-Peter Kriegel, Jörg Sander, Xiaowei Xu, et al. 1996. A density- based algorithm for discovering clusters in large spatial databases with noise. In kdd, Vol. 96. 226–231

    1996

  11. [19]

    Absalom E Ezugwu, Abiodun M Ikotun, Olaide O Oyelade, Laith Abualigah, Jeffery O Agushaka, Christopher I Eke, and Andronicus A Akinyelu. 2022. A com- prehensive survey of clustering algorithms: State-of-the-art machine learning applications, taxonomy, challenges, and future re...

    2022

  12. [20]

    Yoav Freund and Robert E Schapire. 1997. A decision-theoretic generalization of on-line learning and an application to boosting. Journal of computer and system sciences 55, 1 (1997), 119–139

    1997

  13. [21]

    Brendan J Frey and Delbert Dueck. 2007. Clustering by passing messages between data points. science 315, 5814 (2007), 972–976

    2007

  14. [22]

    Jerome H FRIEDMAN and Bogdan E POPESCU. 2008. PREDICTIVE LEARNING VIA RULE ENSEMBLES. The Annals of applied statistics 2, 3 (2008), 916–954

    2008

  15. [23]

    Nave Frost, Michal Moshkovitz, and Cyrus Rashtchian. 2020. ExKMC: Expanding Explainable𝑘-Means Clustering. arXiv preprint arXiv:2006.02399 (2020)

    2020 arXiv

  16. [24]

    Yongjian Fu and Jiawei Han. 1995. Meta-Rule-Guided Mining of Association Rules in Relational Databases.. In KDOOD/TDOOD. Citeseer, 39–46

    1995

  17. [25]

    Mohamed H Gad-Elrab, Daria Stepanova, Trung-Kien Tran, Heike Adel, and Gerhard Weikum. 2020. Excut: Explainable embedding-based clustering over knowledge graphs. In International Semantic Web Conference. Springer, 218–237

    2020

  18. [26]

    Salvador García, Sergio Ramírez-Gallego, Julián Luengo, José Manuel Benítez, and Francisco Herrera. 2016. Big data preprocessing: methods and prospects. Big Data Analytics 1, 1 (2016), 1–22

    2016

  19. [27]

    Riccardo Guidotti, Anna Monreale, Fosca Giannotti, Dino Pedreschi, Salvatore Ruggieri, and Franco Turini. 2019. Factual and counterfactual explanations for black box decision making. IEEE Intelligent Systems 34, 6 (2019), 14–23

    2019

  20. [28]

    Jiawei Han, Jian Pei, and Yiwen Yin. 2000. Mining frequent patterns without candidate generation. ACM sigmod record 29, 2 (2000), 1–12

    2000

  21. [29]

    Harris, K

    Charles R. Harris, K. Jarrod Millman, Stéfan J. van der Walt, Ralf Gommers, Pauli Virtanen, David Cournapeau, Eric Wieser, Julian Taylor, Sebastian Berg, Nathaniel J. Smith, Robert Kern, Matti Picus, Stephan Hoyer, Marten H. van Kerkwijk, Matthew Brett, Allan Haldane, Jaime Fe...

    2020

  22. [30]

    Douglas M Hawkins. 2004. The problem of overfitting. Journal of chemical information and computer sciences 44, 1 (2004), 1–12

    2004

  23. [31]

    Jochen Hipp, Andreas Myka, Rüdiger Wirth, and Ulrich Güntzer. 1998. A new algorithm for faster mining of generalized association rules. In Principles of Data Mining and Knowledge Discovery: Second European Symposium, PKDD’98 Nantes, France, September 23–26, 1998 Proceedings 2 ...

    1998

  24. [32]

    Mark Ibrahim, Melissa Louie, Ceena Modarres, and John Paisley. 2019. Global explanations of neural networks: Mapping the landscape of predictions. In Pro- ceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society . 279–287

    2019

  25. [33]

    Eser Kandogan. 2012. Just-in-time annotation of clusters, outliers, and trends in point-based data visualizations. In 2012 IEEE Conference on Visual Analytics Science and Technology (V AST). IEEE, 73–82

    2012

  26. [34]

    Jacob Kauffmann, Malte Esders, Lukas Ruff, Grégoire Montavon, Wojciech Samek, and Klaus-Robert Müller. 2022. From clustering to cluster explanations via neural networks. IEEE Transactions on Neural Networks and Learning Systems35, 2 (2022), 1926–1940

    2022

  27. [35]

    Daniel Kunkle, Donghui Zhang, and Gene Cooperman. 2008. Mining frequent generalized itemsets and generalized association rules without redundancy.Jour- nal of Computer Science and Technology 23, 1 (2008), 77–102

    2008

  28. [36]

    Bum Chul Kwon, Ben Eysenbach, Janu Verma, Kenney Ng, Christopher De Filippi, Walter F Stewart, and Adam Perer. 2017. Clustervision: Visual supervision of unsupervised clustering. IEEE transactions on visualization and computer graphics 24, 1 (2017), 142–151

    2017

  29. [37]

    Pantelis Linardatos, Vasilis Papastefanopoulos, and Sotiris Kotsiantis. 2021. Ex- plainable ai: A review of machine learning interpretability methods. Entropy 23, 1 (2021), 18

    2021

  30. [38]

    Stuart Lloyd. 1982. Least squares quantization in PCM. IEEE transactions on information theory 28, 2 (1982), 129–137

    1982

  31. [39]

    Wei-Yin Loh. 2011. Classification and regression trees. Wiley interdisciplinary reviews: data mining and knowledge discovery 1, 1 (2011), 14–23

    2011

  32. [40]

    Scott M Lundberg and Su-In Lee. 2017. A unified approach to interpreting model predictions. Advances in neural information processing systems 30 (2017)

    2017

  33. [41]

    Tim Miller. 2019. Explanation in artificial intelligence: Insights from the social sciences. Artificial intelligence 267 (2019), 1–38

    2019

  34. [42]

    Michal Moshkovitz, Sanjoy Dasgupta, Cyrus Rashtchian, and Nave Frost. 2020. Explainable k-means and k-medians clustering. In International conference on machine learning. PMLR, 7055–7065

    2020

  35. [43]

    Daniel Müllner. 2011. Modern hierarchical, agglomerative clustering algorithms. arXiv preprint arXiv:1109.2378 (2011)

    2011 arXiv

  36. [44]

    Guillermo Navas-Palencia. 2020. Optimal binning: mathematical programming formulation. arXiv preprint arXiv:2001.08025 (2020)

    2020 arXiv

  37. [45]

    The pandas development team. 2020. pandas-dev/pandas: Pandas. https://doi. org/10.5281/zenodo.3509134

    2020 doi

  38. [46]

    Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, et al . 2011. Scikit-learn: Machine learning in Python. the Journal of machine Learning research 12 (2011), ...

    2011

  39. [47]

    Why should i trust you?

    Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. " Why should i trust you?" Explaining the predictions of any classifier. InProceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining . 1135–1144

    2016

  40. [48]

    Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2018. Anchors: High- precision model-agnostic explanations. In Proceedings of the AAAI conference on artificial intelligence, Vol. 32

    2018

  41. [49]

    Peter J Rousseeuw. 1987. Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of computational and applied mathematics 20 (1987), 53–65

    1987

  42. [50]

    Mirka Saarela and Susanne Jauhiainen. 2021. Comparison of feature importance measures as explanations for classification models. SN Applied Sciences 3, 2 (2021), 272

    2021

  43. [51]

    Omer Sagi and Lior Rokach. 2018. Ensemble learning: A survey. Wiley interdis- ciplinary reviews: data mining and knowledge discovery 8, 4 (2018), e1249

    2018

  44. [52]

    Vidya Setlur, Michael Correll, and Sarah Battersby. 2022. Oscar: A semantic- based data binning approach. In 2022 IEEE Visualization and Visual Analytics (VIS). IEEE, 100–104

    2022

  45. [53]

    Avanti Shrikumar, Peyton Greenside, and Anshul Kundaje. 2017. Learning important features through propagating activation differences. In International Conference on Machine Learning . PMLR, 3145–3153

    2017

  46. [54]

    Ramakrishnan Srikant and Rakesh Agrawal. 1997. Mining generalized association rules. Future generation computer systems 13, 2-3 (1997), 161–180

    1997

  47. [55]

    Kritsada Sriphaew and Thanaruk Theeramunkong. 2002. A new method for finding generalized frequent itemsets in generalized association rule mining. In Proceedings ISCC 2002 Seventh International Symposium on Computers and Communications. IEEE, 1040–1045. 13

    2002

  48. [56]

    Mukund Sundararajan, Ankur Taly, and Qiqi Yan. 2017. Axiomatic attribution for deep networks. In International Conference on Machine Learning . PMLR, 3319–3328

    2017

  49. [57]

    Sariel Tutay and Amit Somech. 2023. https://github.com/analysis-bots/cluster- explorer

    2023

  50. [58]

    Sariel Tutay and Amit Somech. 2023. Cluster-Explorer: An interactive Framework for Explaining Black-Box Clustering Results. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management . 5106–5110

    2023

  51. [59]

    UC Irvine Machine Learning Repository. 2024. https://archive.ics.uci.edu/

    2024

  52. [60]

    Ulrike Von Luxburg. 2007. A tutorial on spectral clustering. Statistics and computing 17 (2007), 395–416

    2007

  53. [61]

    Danding Wang, Qian Yang, Ashraf Abdul, and Brian Y Lim. 2019. Designing theory-driven user-centric explainable AI. In Proceedings of the 2019 CHI confer- ence on human factors in computing systems . 1–15

    2019

  54. [62]

    Haizhou Wang and Mingzhou Song. 2011. Ckmeans. 1d. dp: optimal k-means clustering in one dimension by dynamic programming. The R journal 3, 2 (2011), 29

    2011

  55. [63]

    Jiazhi Xia, Linquan Huang, Weixing Lin, Xin Zhao, Jing Wu, Yang Chen, Ying Zhao, and Wei Chen. 2022. Interactive visual cluster analysis by contrastive dimensionality reduction. IEEE Transactions on Visualization and Computer Graphics 29, 1 (2022), 734–744

    2022

  56. [64]

    Dongkuan Xu and Yingjie Tian. 2015. A comprehensive survey of clustering algorithms. Annals of Data Science 2, 2 (2015), 165–193

    2015

  57. [65]

    Guizhen Yang. 2004. The complexity of mining maximal frequent itemsets and maximal frequent patterns. InProceedings of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining . 344–353

    2004

  58. [66]

    Xue Ying. 2019. An overview of overfitting and its solutions. InJournal of physics: Conference series, Vol. 1168. IOP Publishing, 022022

    2019

  59. [67]

    Jun Yuan, Changjian Chen, Weikai Yang, Mengchen Liu, Jiazhi Xia, and Shixia Liu. 2021. A survey of visual analytics techniques for machine learning. Compu- tational Visual Media 7 (2021), 3–36

    2021

  60. [68]

    Shichao Zhang, Chengqi Zhang, and Qiang Yang. 2003. Data preparation for data mining. Applied artificial intelligence 17, 5-6 (2003), 375–381

    2003

  61. [69]

    Tian Zhang, Raghu Ramakrishnan, and Miron Livny. 1996. BIRCH: an efficient data clustering method for very large databases. ACM sigmod record 25, 2 (1996), 103–114

    1996

  62. [70]

    Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba

  63. [71]

    Alexander Zien, Nicole Krämer, Sören Sonnenburg, and Gunnar Rätsch. 2009. The feature importance ranking measure. In Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2009, Bled, Slovenia, September 7-11, 2009, Proceedings, Part II 20 . Spri...

    2009

  64. [2016]

    In Proceedings of the IEEE conference on computer vision and pattern recognition

    Learning deep features for discriminative localization. In Proceedings of the IEEE conference on computer vision and pattern recognition . 2921–2929

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.