Learning Model-Agnostic Counterfactual Explanations for Tabular Data
Open accessThe DOI matches its Crossref record; 29 of the 33 references it cites were found in a registry or an index.
Abstract
Counterfactual explanations can be obtained by identifying the smallest change made to an input vector to influence a prediction in a positive way from a user’s viewpoint; for example, from ’loan rejected’ to ’awarded’ or from ’high risk of cardiovascular disease’ to ’low risk’. Previous approaches would not ensure that the produced counterfactuals be proximate (i.e., not local outliers) and connected to regions with substantial data density (i.e., close to correctly classified observations), two requirements known as counterfactual faithfulness. Our contribution is twofold. First, drawing ideas from the manifold learning literature, we develop a framework, called C-CHVAE, that generates faithful counterfactuals. Second, we suggest to complement the catalog of counterfactual quality measures using a criterion to quantify the degree of difficulty for a certain counterfactual suggestion. Our real world experiments suggest that faithful counterfactuals come at the cost of higher degrees of difficulty.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The first method is Pragmatism, which quantifies, based on domain knowledge, the degree of difficulty for an individual to act upon the suggestions in a counterfactual explanation [187,203].”
From From Anecdotal Evidence to Quantitative Evaluation Methods: A Systematic Review on Evaluating Explainable AI (2023) · cites this paper as [187]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
(1996) · Cited by 20,179
Auto-Encoding Variational Bayes
Kingma, D. P., & Welling, M. (2013) · arXiv · doi:10.48550/arxiv.1312.6114 · Cited by 20,067 · cited 2×
“We suggest embedding counterfactual search into a data density approximator, here a variational autoencoder (VAE) [9].”
BackgroundMethodLOF: identifying density-based local outliers
Breunig, M. M., Kriegel, H.-P., Ng, R. T., & Sander, J. (2000) · ACM SIGMOD Record · doi:10.1145/335191.335388 · Cited by 5,207
Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey
Akhtar, N., & Mian, A. (2018) · IEEE Access · doi:10.1109/access.2018.2807385 · Cited by 2,021
Makhzani, A., Shlens, J., Jaitly, N., Goodfellow, I., & Frey, B. (2015) · arXiv · doi:10.48550/arxiv.1511.05644 · Cited by 1,579
A Reductions Approach to Fair Classification
Agarwal, A., Beygelzimer, A., Dudík, M., Langford, J., & Wallach, H. (2018) · arXiv · doi:10.48550/arxiv.1803.02453 · Cited by 1,288 · cited 2×
“We generate x = [x1,x2] from a mixture of 3 Gaussians with μ = [μ1, μ2, μ3] and σ = [σ1,σ2,σ3] = [1, 1, 1] with a fixed seed.”
MethodCounterfactual explanations without opening the black box: automated decisions and the GDPR
Wachter, S., Mittelstadt, B., & Russell, C. (2017) · SSRN Electronic Journal · doi:10.2139/ssrn.3063289 · Cited by 1,061
Auto-Encoding Variational Bayes
Chen, Y., Liu, J., Peng, L., Wu, Y., Xu, Y., & Zhang, Z. (2024) · Cambridge Explorations in Arts and Sciences · doi:10.61603/ceas.v2i1.33 · Cited by 934
Tolstikhin, I., Bousquet, O., Gelly, S., & Schoelkopf, B. (2017) · arXiv · doi:10.48550/arxiv.1711.01558 · Cited by 695
Zafar, M. B., Valera, I., Gomez Rodriguez, M., & Gummadi, K. P. (2017) · Proceedings of the 26th International Conference on World Wide Web · doi:10.1145/3038912.3052660 · Cited by 632 · cited 1×
“We take the pretrained, potentially non-linear, classifier f (·) as given, which can also be a training time fairness constraint classifier [1, 22].”
MethodTolstikhin, I., Bousquet, O., Gelly, S., & Schölkopf, B. (2018) · MPG.PuRe (Max Planck Society) · Cited by 602
Sex and gender differences in health: Science & Society Series on Sex and Science
Regitz‐Zagrosek, V. (2012) · The EMBO Reports · doi:10.1038/embor.2012.87 · Cited by 572
Brown, T. B., Mané, D., Roy, A., Abadi, M., & Gilmer, J. (2017) · arXiv · doi:10.48550/arxiv.1712.09665 · Cited by 509
Actionable Recourse in Linear Classification
Ustun, B., Spangher, A., & Liu, Y. (2019) · Proceedings of the Conference on Fairness, Accountability, and Transparency · doi:10.1145/3287560.3287566 · Cited by 501 · cited 5×
“Both data sets are standard in the literature [4, 16, 20] and are described in more detail in our github repository.”
BackgroundMethodInfluentialHandling incomplete heterogeneous data using VAEs
Nazábal, A., Olmos, P. M., Ghahramani, Z., & Valera, I. (2020) · Pattern Recognition · doi:10.1016/j.patcog.2020.107501 · Cited by 345
Efficient Search for Diverse Coherent Explanations
Russell, C. (2019) · Proceedings of the Conference on Fairness, Accountability, and Transparency · doi:10.1145/3287560.3287569 · Cited by 218 · cited 2×
“Both data sets are standard in the literature [4, 16, 20] and are described in more detail in our github repository.”
BackgroundGrgic-Hlaca, N., Redmiles, E. M., Gummadi, K. P., & Weller, A. (2018) · Proceedings of the 2018 World Wide Web Conference on World Wide Web - WWW '18 · doi:10.1145/3178876.3186138 · Cited by 207 · cited 1×
“Furthermore, existing methods leave the elicitation of appropriate distance/cost functions up to (expert) opinions [4, 10, 12, 21, 22], which can vary considerably across individuals [5].”
MethodInterpretable Predictions of Tree-based Ensembles via Actionable Feature Tweaking
Tolomei, G., Silvestri, F., Haines, A., & Lalmas, M. (2017) · Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/3097983.3098039 · Cited by 129 · cited 2×
“[18] assume the pretrained classifiers to be linear or tree based respectively, which can restrict usefulness.”
BackgroundSex and Gender Differences in Pharmacology
Regitz‐Zagrosek, V. (2012) · Handbook of experimental pharmacology · doi:10.1007/978-3-642-30726-3 · Cited by 125
Variational Autoencoder with Arbitrary ConditioningPreprint
Ivanov, O., Figurnov, M., & Vetrov, D. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1806.02382 · Cited by 107 · cited 1×
“It can also be combined with a multitude of autoencoder architectures [7, 8, 12, 13, 17, 19]; • No ad-hoc distance measures for input data.”
BackgroundGhojogh, B., Crowley, M., Karray, F. O., & Ghodsi, A. (2022) · Springer eBooks · doi:10.1007/978-3-031-10602-6_21 · Cited by 106
Joshi, S. D., Koyejo, S., Vijitbenjaronk, W. D., Kim, B., & Ghosh, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1907.09615 · Cited by 90 · cited 1×
“[8] suggest a similar explanation model, however, they focus on causal models and are less concerned with the issue of evaluating counterfactual explanations.”
BackgroundInterpretable Credit Application Predictions With Counterfactual Explanations
McGrath, R., Costabello, L., Van, C. L., Sweeney, P., Kamiab, F., Shen, Z., & Lécué, F. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1811.05245 · Cited by 78 · cited 1×
“nce in a meaningful way (e. g. measuring distance between different occupations). Furthermore, existing methods leave the elicitation of appropriate distance (or cost functions) up to expert opinions [2, 3, 4, 5, 6], which can vary considerably across individuals [7]. Therefore, we suggest measuring similarity between…”
BackgroundGrgić-Hlača, N., Redmiles, E. M., Gummadi, K. P., & Weller, A. (2018) · arXiv · doi:10.48550/arxiv.1802.09548 · Cited by 75 · cited 1×
“Furthermore, existing methods leave the elicitation of appropriate distance/cost functions up to (expert) opinions [4, 10, 12, 21, 22], which can vary considerably across individuals [5].”
MethodOn decompositional algorithms for uniform sampling from n -spheres and n -balls
Harman, R., & Lacko, V. (2010) · Journal of Multivariate Analysis · doi:10.1016/j.jmva.2010.06.002 · Cited by 73 · cited 1×
“The details can be found in our github repository, but it uses a standard procedure to generate random numbers distributed uniformly over a sphere [6, 12] around the latent observation ẑ.”
MethodIssues with post-hoc counterfactual explanations: a discussionPreprint
Laugel, T., Lesot, M., Marsala, C., & Detyniecki, M. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.04774 · Cited by 65 · cited 4×
“al counterfactual found by a method. As an example, a cost of p suggests changing a free feature by at least ppercentiles to receive a desired result. More particularly, the two measures suggested in [1] and shortly described above measure proximity and connectedness of counterfactuals, but do not indicate the degree…”
BackgroundMethodInfluentialGeneralized Inverse Classification
Lash, M. T., Lin, Q., Street, N., Robinson, J. G., & Ohlmann, J. (2017) · Proceedings of the 2017 SIAM International Conference on Data Mining · doi:10.1137/1.9781611974973.19 · Cited by 57 · cited 7×
“While GS works for different classifiers, the AR and HCLS algorithms do not.”
BackgroundMethodInfluentialInverse Classification for Comparison-based Interpretability in Machine Learning
Laugel, T., Lesot, M., Marsala, C., Renard, X., & Detyniecki, M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1712.08443 · Cited by 44 · cited 4×
“Furthermore, existing methods leave the elicitation of appropriate distance/cost functions up to (expert) opinions [4, 10, 12, 21, 22], which can vary considerably across individuals [5].”
BackgroundMethodInfluentialTowards User Empowerment.Preprint
Pawelczyk, M., Haug, J., Broelemann, K., & Kasneci, G. (2019) · arXiv (Cornell University) · Cited by 6
Self-citation
Research works printed but not linked: 1
- [18] Kihyuk Sohn, Honglak Lee, and Xinchen Yan. 2015. Learning structured output representation using deep conditional generative models. In Advances in neural information processing systems . 3483-3491. -
Cited by
Deep Neural Networks and Tabular Data: A Survey
Borisov, V., Leemann, T., Seßler, K., Haug, J., Pawelczyk, M., & Kasneci, G. (2024) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3229161 · Cited by 1,025
“To alleviate this problem and to suggest realistic looking counterfactuals, researchers have suggested building recourse suggestions on generative models [179]–[184].”
BackgroundSelf-citationNauta, M., Trienes, J., Pathak, S., Nguyen, E., Peters, M., Schmitt, Y., Schlötterer, J., van Keulen, M., & Seifert, C. (2023) · ACM Computing Surveys · doi:10.1145/3583558 · Cited by 590
Counterfactual explanations and how to find them: literature review and benchmarking
Guidotti, R. (2024) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-022-00831-6 · Cited by 471
Stepin, I., Alonso, J. M., Catala, A., & Pereira-Farina, M. (2021) · IEEE Access · doi:10.1109/access.2021.3051315 · Cited by 376
“make use of an autoencoder architecture for a pretrained classifier performing counterfactual search in the nearest neighbor style [151].”
BackgroundMethodInfluentialChou, Y.-L., Moreira, C., Bruza, P., Ouyang, C., & Jorge, J. (2022) · Information Fusion · doi:10.1016/j.inffus.2021.11.003 · Cited by 302
Metadata
- Authors
- Pawelczyk, M., Broelemann, K., & Kasneci, G.
- Affiliations
- University of Tübingen (DE)
- Schufa Holding AG, Germany
- Venue
- Proceedings of The Web Conference 2020
- Publisher
- ACM
- Published
- 2020-04-20
- Pages
- 3126-3132
- OpenAlex
- W3023497337
- Cited by
- 167
- References
- 33
- Citations by year
- 2026: 21 · 2025: 24 · 2024: 34 · 2023: 34 · 2022: 19 · 2021: 29 · 2020: 6
- FWCI
- 8.39
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- International World Wide Web Conference (WWW) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Keywords
- counterfactual explanations, tabular data, manifold learning, model-agnostic explanation, explainable machine learning
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.