Abstract
This work introduces interpretable regional descriptors, or IRDs, for local, model-agnostic interpretations. IRDs are hyperboxes that describe how an observation’s feature values can be changed without affecting its prediction. They justify a prediction by providing a set of “even if” arguments (semi-factual explanations), and they indicate which features affect a prediction and whether pointwise biases or implausibilities exist. A concrete use case shows that this is valuable for both machine learning modelers and persons subject to a decision. We formalize the search for IRDs as an optimization problem and introduce a unifying framework for computing IRDs that covers desiderata, initialization techniques, and a post-processing method. We show how existing hyperbox methods can be adapted to fit into this unified framework. A benchmark study compares the methods based on several quality measures and identifies two strategies to improve IRDs.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
In contrast, a counterfactual would be “if you had rich savings and become highly skilled, your credit would be a low risk”. Such statements are not covered by IRDs.
- 2.
However, the concrete strategies can only reveal counterfactual explanations [31].
- 3.
Note that if all genders are part of the box, it does not mean the model is fair.
- 4.
- 5.
- 6.
For classification models, \(Y' \subset [0, 1]\) must hold.
- 7.
For this, we extended the optimization task of Ribeiro et al. [25] to target IRDs by aiming for a precision of 1 and by including the locality constraint.
- 8.
Double-in-size refers to the size of the training data, not of \(\bar{\underline{{\textbf {X}}}}\).
- 9.
We prefer this measure over computation time because it is independent of the concrete implementation. We have made our best efforts to implement the methods efficiently, but there is usually room for improvement.
- 10.
These data points can also be excluded from the data before training a model. However, our experiments showed the results for the RQs are almost the same.
- 11.
The true hyperbox of the CART model might be larger than the terminal node-induced hyperbox (see Figure S. 5 in the Appendix).
- 12.
The size decuples instead of doubles compared to the training data, because not all training data are \( \in \bar{\underline{B}}\) and, thus, not in \(\bar{\underline{{\textbf {X}}}}\).
References
Dash, S., Günlük, O., Wei, D.: Boolean decision rules via column generation. In: Proceedings of the 32nd International Conference on Neural Information Processing Systems, NIPS 2018, pp. 4660–4670. Curran Associates Inc., Red Hook, NY, USA (2018)
Dhurandhar, A., et al.: Explanations based on the missing: towards contrastive explanations with pertinent negatives. In: Proceedings of the 32nd International Conference on Neural Information Processing Systems, NIPS 2018, pp. 590–601. Curran Associates Inc., Red Hook, NY, USA (2018)
Doshi-Velez, F., Kim, B.: Towards a rigorous science of interpretable machine learning. arXiv 1702.08608 v2, arXiv.org E-Print Archive (2017). 10.48550/arXiv. 1702.08608
Dua, D., Graff, C.: UCI machine learning repository (2017). www.archive.ics.uci.edu/ml/datasets/statlog+(german+credit+data)
Eckstein, J., Hammer, P.L., Liu, Y., Nediak, M., Simeone, B.: The maximum box problem and its application to data analysis. Comput. Optim. Appl. 23(3), 285–298 (2002). https://doi.org/10.1023/a:1020546910706
El Shawi, R., Sherif, Y., Al-Mallah, M., Sakr, S.: Interpretability in healthcare: a comparative study of local machine learning interpretability techniques. Comput. Intell. 37(4), 1633–1650 (2021). https://doi.org/10.1111/coin.12410
Emmerich, M.T.M., Deutz, A.H., Kruisselbrink, J.W.: On quality indicators for black-box level set approximation. In: Tantar, E., et al. (eds.) EVOLVE- A Bridge between Probability, Set Oriented Numerics and Evolutionary Computation, pp. 157–185. Springer, Berlin (2013). https://doi.org/10.1007/978-3-642-32726-1_4
Fan, M., Wei, W., Xie, X., Liu, Y., Guan, X., Liu, T.: Can we trust your explanations? Sanity checks for interpreters in android malware analysis. IEEE Tran. Inf. Forensics Secur. 16, 838–853 (2021). https://doi.org/10.1109/TIFS.2020.3021924
Fernandez, G., Aledo, J.A., Gamez, J.A., Puerta, J.M.: Factual and counterfactual explanations in fuzzy classification trees. IEEE Trans. Fuzzy Syst. 30(12), 5484–5495 (2022). https://doi.org/10.1109/tfuzz.2022.3179582
Ferreira, L.: German credit risk (2018). www.kaggle.com/datasets/kabure/german-credit-data-with-risk. Accessed 23 Jan 2023
Friedman, J.H., Fisher, N.I.: Bump hunting in high-dimensional data. Stat. Comput. 9(2), 123–143 (1999). https://doi.org/10.1023/A:1008894516817
Fürnkranz, J., Kliegr, T.: A brief overview of rule learning. In: Bassiliades, N., Gottlob, G., Sadri, F., Paschke, A., Roman, D. (eds.) RuleML 2015. LNCS, vol. 9202, pp. 54–69. Springer, Cham (2015). https://doi.org/10.1007/978-3-319-21542-6_4
Goldstein, A., Kapelner, A., Bleich, J., Pitkin, E.: Peeking inside the black box: Visualizing statistical learning with plots of individual conditional expectation. J. Comput. Graph. Stat. 24(1), 44–65 (2015). https://doi.org/10.1080/10618600.2014.907095
Guidotti, R., Monreale, A., Giannotti, F., Pedreschi, D., Ruggieri, S., Turini, F.: Factual and counterfactual explanations for black box decision making. IEEE Intell. Syst. 34(6), 14–23 (2019). https://doi.org/10.1109/MIS.2019.2957223
Guidotti, R., Monreale, A., Ruggieri, S., Naretto, F., Turini, F., Pedreschi, D., Giannotti, F.: Stable and actionable explanations of black-box models through factual and counterfactual rules. Data Min. Knowl. Disc. (2022). https://doi.org/10.1007/s10618-022-00878-5
Guidotti, R., Monreale, A., Ruggieri, S., Pedreschi, D., Turini, F., Giannotti, F.: Local rule-based explanations of black box decision systems. arXiv 1805.10820, arXiv.org E-Print Archive (2018). 10.48550/arXiv. 1805.10820
Kenny, E.M., Keane, M.T.: On generating plausible counterfactual and semi-factual explanations for deep learning. Proc. AAAI Conf. Artif. Intell. 35(13), 11575–11585 (2021). https://doi.org/10.1609/aaai.v35i13.17377
Khuat, T.T., Ruta, D., Gabrys, B.: Hyperbox-based machine learning algorithms: a comprehensive survey. Soft Comput. 25(2), 1325–1363 (2020). https://doi.org/10.1007/s00500-020-05226-7
Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization. arXiv 1412.6980 v9, arXiv.org E-Print Archive (2017). 10.48550/arXiv. 1412.6980
Kuratomi, A., Miliou, I., Lee, Z., Lindgren, T., Papapetrou, P.: JUICE: JUstIfied counterfactual explanations. In: Pascal, P., Ienco, D. (eds.) Discovery Science. pp. 493–508. LNCS, Springer, Cham (2022). https://doi.org/10.1007/978-3-031-18840-4_35
Land, A.H., Doig, A.G.: An automatic method of solving discrete programming problems. Econometrica 28(3), 497–520 (1960). https://doi.org/10.2307/1910129
Lemhadri, I., Li, H.H., Hastie, T.: RbX: region-based explanations of prediction models. arXiv 2210.08721, arXiv.org E-Print Archive (2022). 10.48550/arXiv.2210.08721
Nugent, C., Doyle, D., Cunningham, P.: Gaining insight through case-based explanation. J. Intell. Inf. Syst. 32(3), 267–295 (2009). https://doi.org/10.1007/s10844-008-0069-0
Ribeiro, M.T., Singh, S., Guestrin, C.: Why should I trust you? Explaining the predictions of any classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 1135–1144 (2016)
Ribeiro, M.T., Singh, S., Guestrin, C.: Anchors: High-precision model-agnostic explanations. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 32, no. 1 (2018). https://doi.org/10.1609/aaai.v32i1.11491
Ribeiro, M.T., Singh, S., Guestrin, C.: Anchor. Github repository. www.github.com/marcotcr/anchor (2022), Commit: b1f5e6ca37428613723597e85c38558e8cd21c2e
Schwartzenberg, C., van Engers, T.M., Li, Y.: The fidelity of global surrogates in interpretable machine learning. BNAIC/BeneLearn 2020 (2020)
Sharma, R., Reddy, N., Kamakshi, V., Krishnan, N.C., Jain, S.: MAIRE - a model-agnostic interpretable rule extraction procedure for explaining classifiers. In: Holzinger, A., Kieseberg, P., Tjoa, A.M., Weippl, E. (eds.) CD-MAKE 2021. LNCS, vol. 12844, pp. 329–349. Springer, Cham (2021). https://doi.org/10.1007/978-3-030-84060-0_21
Stepin, I., Alonso, J.M., Catala, A., Pereira-Fariña, M.: Generation and evaluation of factual and counterfactual explanations for decision trees and fuzzy rule-based classifiers. In: 2020 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), pp. 1–8. IEEE, Glasgow, United Kingdom (2020). https://doi.org/10.1109/FUZZ48607.2020.9177629
Vanschoren, J., van Rijn, J.N., Bischl, B., Torgo, L.: OpenML: networked science in machine learning. SIGKDD Explor. Newsl. 15(2), 49–60 (2014). https://doi.org/10.1145/2641190.2641198
Wachter, S., Mittelstadt, B., Russell, C.: Counterfactual explanations without opening the black box: automated decisions and the GDPR. Harvard J. Law Technol. 31(2), 841–887 (2018)
Zabinsky, Z.B., Huang, H.: A partition-based optimization approach for level set approximation: probabilistic branch and bound. In: Smith, A.E. (ed.) Women in Industrial and Systems Engineering. WES, pp. 113–155. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-11866-2_6
Zabinsky, Z.B., Wang, W., Prasetio, Y., Ghate, A., Yen, J.W.: Adaptive probabilistic branch and bound for level set approximation. In: Proceedings of the 2011 Winter Simulation Conference (WSC), pp. 4146–4157. IEEE, Phoenix, AZ, USA (2011). https://doi.org/10.1109/WSC.2011.6148103
Acknowledgements
This work has been partially supported by the Federal Statistical Office of Germany.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Ethics declarations
Ethical Statement
For this work, no personal data was collected or processed. Only open source datasets were used for the illustrative example and the benchmark study. Furthermore, our work does not aim at a possible use for policing or military.
Rights and permissions
Copyright information
© 2023 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Dandl, S., Casalicchio, G., Bischl, B., Bothmann, L. (2023). Interpretable Regional Descriptors: Hyperbox-Based Local Explanations. In: Koutra, D., Plant, C., Gomez Rodriguez, M., Baralis, E., Bonchi, F. (eds) Machine Learning and Knowledge Discovery in Databases: Research Track. ECML PKDD 2023. Lecture Notes in Computer Science(), vol 14171. Springer, Cham. https://doi.org/10.1007/978-3-031-43418-1_29
Download citation
DOI: https://doi.org/10.1007/978-3-031-43418-1_29
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-43417-4
Online ISBN: 978-3-031-43418-1
eBook Packages: Computer ScienceComputer Science (R0)
