Pith. sign in

REVIEW 3 major objections 3 minor 73 references

Transparency, Auditability and eXplainability of Machine Learning Models in Credit Scoring

T0 review · 3 major / 3 minor · reviewed 2026-08-27 · deepseek-v4-flash

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

Pith's one-line read A structured explainability framework can give black-box credit-scoring models the same audit-readiness as logistic scorecards, and a HELOC case study shows how.

desk verdict TAX4CS is a useful organizing framework for credit-scoring XAI, but the case study leans on visual inspection where it claims auditability; worth refereeing with revisions. read the letter →

arxiv 2009.13384 v1 pith:KHX42N5U submitted 2020-09-28 stat.ML cs.LGecon.GNq-fin.ECstat.APstat.ME

Michael Bücker , Gero Szepannek , Alicja Gosiewska , Przemyslaw Biecek This is my paper · ORCID
classification stat.MLcs.LGecon.GNq-fin.ECstat.APstat.ME
keywords creditscoringmachinelearninginterpretabilityexplainableAImodel-agnosticexplanationspartialdependenceplotsSHAPvaluesscorecardmodellingregulatoryauditability
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 proposes TAX4CS, a structured framework for making "black box" machine-learning credit-scoring models transparent, auditable, and explainable, so that banks could use more accurate ML models without losing the regulatory readability of logistic scorecards. It argues that transparency has several distinct dimensions—global versus local, and model-level versus instance-level—and that each regulatory requirement can be mapped to a concrete model-agnostic explanation technique. A real-world case study on FICO HELOC data demonstrates the workflow on a scorecard, a gradient boosting machine, a support vector machine, and a spline-based logistic regression. The paper's central finding is that these ML models can be interpreted at a level comparable to traditional scorecards, while on this data the scorecard itself remains a strong performer, so model complexity should be justified by benchmarking rather than assumed superior.

What carries the argument

The TAX4CS framework is the carrying mechanism: a four-part process that first identifies stakeholders, then maps their involvement across the model's lifetime, then lists their regulatory needs, and finally selects XAI methods from a pyramid. The pyramid's two dimensions are depth (performance measures, then variable importance, then variable-effect profiles) and audience (global explanations of overall model behaviour versus local explanations of individual predictions). The work of the framework is to translate legal and supervisory demands—right to explanation, auditability, and monitoring—into concrete model-agnostic tools: permutation-based feature importance, partial dependence plots, additive breakdown attributions such as iBreakDown or SHAP, and ceteris paribus profiles, all applicable to any predictive model.

What would settle it

Train a black-box model on a synthetic credit dataset with a known interaction between two variables, run the TAX4CS workflow, and compare the partial dependence and ceteris paribus outputs with the model's true counterfactual responses and with exact Shapley values; if the explanation outputs misstate the effect direction for a material share of applicants, the framework's auditability claim is falsified.

Watch

Extended reading notes

Core claim

The central discovery the paper argues for is that the interpretability gap between traditional scorecards and complex machine-learning models is not inherent: with the right structured process, black-box credit models can be understood at both the model level and the level of a single credit decision. The proposed TAX4CS framework selects explanation methods by stakeholder needs and organizes them in a pyramid whose vertical axis is depth of exploration (overall performance, then variable importance, then variable effects) and whose horizontal axis is global versus local explanation. Applying it to the HELOC dataset yields interpretations for gradient boosting and SVM models—permutation importances, partial dependence profiles, breakdown attributions, and ceteris paribus what-if profiles—that are comparable to the point-based explanation a scorecard gives. The paper also finds that, at least for this tabular credit dataset, the simpler scorecard and spline-based logistic regression match or exceed the more complex algorithms, so the framework's value is as much about auditing and comparison as about unlocking accuracy gains.

Load-bearing premise

The framework assumes that model-agnostic explanations—permutation importance, partial dependence, and Shapley-style attributions—faithfully represent what the black-box model actually does, closely enough for auditors and regulators; the paper itself acknowledges partial dependence can diverge from true predictions and that iBreakDown attributions depend on feature ordering.

Editorial extensions

If this is right

  • Banks can use TAX4CS to assemble a regulatory compliance dossier for a complex ML model, covering both model-level monitoring and explanation of individual credit decisions.
  • Champion-challenger comparisons can now include explanation quality, not just AUC, because the same XAI toolkit applies to scorecards and black-box models alike.
  • On tabular credit data, complex machine learning may deliver only a small accuracy gain, so the paper's benchmark supports testing complexity against a scorecard baseline before deployment.
  • If credit scoring moves to richer, harder-to-engineer data sources such as transaction-level records, the framework's structured exploration becomes the route to keeping ML accuracy while satisfying audit expectations.

Reading between the lines

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

  • The framework treats transparency as a set of procedures; it does not certify that the explanations are causally correct, so a stricter reading would require fidelity checks before regulators rely on them.
  • Because partial dependence averages over other features, a direct extension is to flag interactions by measuring how far individual ceteris paribus profiles diverge from the average, turning the framework's own plots into a non-additivity diagnostic.
  • The same pyramid could be applied to fairness audits: replacing performance with a bias metric and drilling down to variable importance would show which drivers carry disparate impact, though causal identification would still need external assumptions.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. The paper proposes TAX4CS, a stakeholder-oriented framework for making credit scoring models transparent, auditable, and explainable, and demonstrates it on the FICO HELOC dataset by comparing a traditional scorecard with several machine learning models (GBM, SVM, spline-based logistic regression, random forests, etc.). The framework structures model-level exploration (performance, permutation importance, partial dependence plots) and instance-level exploration (prediction, iBreakDown/SHAP attributions, ceteris paribus profiles). The authors claim that modern complex machine learning techniques can achieve interpretability comparable to that of traditional logistic regression scorecards while retaining superior predictive power, and they provide reproducible code in a public GitHub repository.

Significance. If the central claim holds, the framework could help banks leverage more accurate machine learning models while still meeting regulatory transparency and auditability requirements, which is a practically important contribution. The paper is useful as a structured compilation of model-agnostic explanation techniques applied to a real credit scoring problem, with reproducible code and a public dataset. Its main empirical evidence is visual and descriptive, however, and the auditability claim is not backed by quantitative fidelity checks, so the significance is conditional on strengthening that evidence.

major comments (3)
  1. [§4.1.3 and §4.2.2] The manuscript acknowledges that partial dependence profiles can differ from true model predictions (citing Szepannek 2019 for a quantitative measure) and that iBreakDown attributions depend on feature ordering, but it does not apply any quantitative fidelity or stability measure in the case study. Since the central claim of "interpretations comparable to traditional logistic regression" (Section 5) rests on these explanations being faithful representations of the black-box models, the absence of such a check leaves the claim supported only by visual inspection of the PDP and attribution plots for a single dataset and split. Please add a quantitative fidelity measure (e.g., Szepannek's) or stability diagnostics for the attributions, or alternatively explicitly scope the claim to the availability of explanations rather than their faithfulness.
  2. [§3.1 and Figure 3/4] The performance comparison is based on a single 75/25 train/test split with no confidence intervals, bootstrap, or significance tests on the AUC values. The conclusion that complex machine learning models show only a minor advantage over the scorecard and that the scorecard is competitive is load-bearing for the paper's recommendation to benchmark models per situation. Without uncertainty quantification, observed AUC differences may be within sampling noise. Please add bootstrap confidence intervals or a resampling-based comparison, and report the dispersion of the performance estimates.
  3. [§3.2.2 and Table 3] The best-performing model, rms_13_tuned, is described as "manually tuned" but its penalty parameters are not reported, and the exact spline transformation specification for rms_13 is not given. This undermines the reproducibility of the main result even though code is available in a GitHub repository. Please report the penalty parameters and the precise model specification (e.g., number and location of knots, type of spline basis) in the appendix.
minor comments (3)
  1. [Throughout] There are several typographical errors, including "Finacial" (Introduction), "ehtical" (Introduction), "detaied" (§2.1), "the the" (§4.3), "suprisingly" (§5), and "explanability" (§5). A proofreading pass is needed.
  2. [Figure 6] The caption of Figure 6 says "Drop-out loss of AUC as measure of variable importance," but the text in §4.1.2 states the variable importance is measured as an increase in 1-AUC after permutation; the caption could be more precise about the direction of the effect.
  3. [§2.2.4] The description of the "pyramid of XAI methods" is conceptual and would benefit from a concrete example of how a stakeholder need maps to a specific method and analysis step in the case study.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: TAX4CS is a proposed workflow evaluated on external data.

full rationale

This paper is an application and position paper rather than a derivation chain. It proposes TAX4CS, a structured workflow that selects existing model-agnostic XAI techniques (permutation importance, PDP, CP profiles, SHAP/iBreakDown) and demonstrates them on the external FICO HELOC data set. There is no fitted parameter subsequently reported as a prediction, no definition that presupposes the claimed conclusion, and no uniqueness theorem or ansatz imported from the authors' prior work. The paper's self-citations (Biecek 2018 for DALEX, Gosiewska and Biecek 2019 for iBreakDown, Szepannek 2017a for scorecard development, Szepannek 2019 for a PDP-fidelity measure) are used as background or implementation references. Importantly, the Szepannek 2019 fidelity measure is mentioned but not applied, so the auditability claim rests on visual inspection rather than on a circular invocation of that measure; this is a correctness limitation, not circularity. The central interpretability comparison is supported by an external benchmark and reproducible code, not by the paper's own fitted values, so there is no self-definitional or fitted-input-called-prediction pattern.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The empirical demonstration depends on a set of tuned hyperparameters and manual scorecard choices; the framework's central guarantee additionally depends on the faithfulness of post hoc explanations and on the assumption that post hoc explanation suffices for regulatory transparency. No new physical or mathematical entity is introduced.

free parameters (3)
  • Scorecard manual binning and selection thresholds = MIV threshold 0.01; minimum relative IV improvement 5%; manually merged bins
    The scorecard baseline is built through manual coarse classing and monotonicity-constrained merging (Section 3.2.1). The exact bin edges and thresholds affect the AUC comparison and the paper's conclusion that a simple model is competitive.
  • Challenger model hyperparameters (gbm, svm, glmnet, xgboost, randomForest, ranger) = e.g., gbm n.trees=10000; svm cost=0.201, gamma=0.019; glmnet alpha=0.388, lambda=0.00112; xgboost eta=0.158…
    Random search tuning on the training split (Section 3.2.2) sets the complexity of every challenger; different tuning could alter the performance ranking that supports the claim about machine learning's predictive advantage.
  • RMS spline and penalty parameters = not reported (manually tuned)
    The best test-AUC model 'rms_13_tuned' is manually tuned, but the penalty and spline parameters are not reported, so the top result cannot be independently reconstructed from the paper.
assumptions (4)
  • domain assumption Model-agnostic XAI methods faithfully represent the behavior of the black-box model
    The framework's guarantee of transparency and auditability assumes that permutation importance, PDPs, SHAP/iBreakDown and ceteris paribus profiles are reliable explanations. Section 2.2.4 selects these methods and Section 4 uses them as evidence. The paper acknowledges deviations, such as PDPs differing from true predictions and iBreakDown being order-dependent.
  • domain assumption Post hoc explanation is sufficient to satisfy regulatory transparency requirements
    TAX4CS is built on the premise that explaining a complex model after training meets GDPR and BCBS-style auditability demands. This is contested by Rudin (2019), which the paper cites in Section 4.3 without resolving the tension.
  • domain assumption A single 75/25 random split and AUC provide an adequate basis for model comparison
    Section 3.1 splits the HELOC data once and Section 4.1.1 restricts the performance comparison to AUC on that split, with no resampling or confidence intervals. The empirical conclusion rests on this assumption.
  • standard math Standard machine learning assumptions about tabular data and loss functions
    Model fitting and AUC calculation use standard statistical learning definitions; these are unproblematic background assumptions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transparency, Auditability and eXplainability of Machine Learning Models in Credit Scoring." pith.science (2026). https://pith.science/paper/KHX42N5U

@misc{pith2026200913384,
  author       = {Pith},
  title        = {Pith review of: Transparency, Auditability and eXplainability of Machine Learning Models in Credit Scoring},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KHX42N5U}},
  note         = {Machine review of arXiv:2009.13384}
}
read the original abstract

A major requirement for credit scoring models is to provide a maximally accurate risk prediction. Additionally, regulators demand these models to be transparent and auditable. Thus, in credit scoring, very simple predictive models such as logistic regression or decision trees are still widely used and the superior predictive power of modern machine learning algorithms cannot be fully leveraged. Significant potential is therefore missed, leading to higher reserves or more credit defaults. This paper works out different dimensions that have to be considered for making credit scoring models understandable and presents a framework for making ``black box'' machine learning models transparent, auditable and explainable. Following this framework, we present an overview of techniques, demonstrate how they can be applied in credit scoring and how results compare to the interpretability of score cards. A real world case study shows that a comparable degree of interpretability can be achieved while machine learning techniques keep their ability to improve predictive power.

Figures

Figures reproduced from arXiv: 2009.13384 by the authors.

Figure 1
Figure 1. The process description consists of four components. The first defines [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Example: automatic vs. manual binning for the variable months since [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Model performance measured by AUC on test data [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Selection of models for comparison based on AUC on training and [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Range of scorecard points as measure of variable importance for the [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Drop-out loss of AUC as measure of variable importance for the [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Marginal effect of variable ’ExternalRiskEstimate’ based on Score [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Ceteris Paribus plots for the variable ’ExternalRiskEstimate’ based [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Partial Dependence plots for the variable ’ExternalRiskEstimate’ [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: ScorecardPoints for a single prediction as individual explanation of [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]
Figure 11
Figure 11. Figure 11: Additive breakdown for a single prediction as individual model [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]
Figure 12
Figure 12. Figure 12: Average additive breakdown aka SHAP values for a single prediction [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 63 canonical work pages

  1. [1]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

  2. [2]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

  3. [3]

    6DTpDQ 2(C

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

    work page 2018

  4. [4]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...

  5. [5]

    Alemzadeh, H., Raman, J., Leveson, N., Kalbarczyk, Z., and Iyer, R. K. (2016). Adverse events in robotic surgery: A retrospective study of 14 years of fda data. PLOS ONE , 11(4):1--20

    work page 2016

  6. [6]

    J., Arroyo , J., Caparrini , A., and Segovia-Vargas , M

    Ariza-Garzón , M. J., Arroyo , J., Caparrini , A., and Segovia-Vargas , M. (2020). Explainability of a machine learning granting scoring model in peer-to-peer lending. IEEE Access , 8:64873--64890

    work page 2020

  7. [7]

    Arya, V., Bellamy, R. K. E., Chen, P.-Y., Dhurandhar, A., Hind, M., Hoffman, S. C., Houde, S., Liao, Q. V., Luss, R., Mojsilović, A., Mourad, S., Pedemonte, P., Raghavendra, R., Richards, J., Sattigeri, P., Shanmugam, K., Singh, M., Varshney, K. R., Wei, D., and Zhang, Y. (2019). One explanation does not fit all: A toolkit and taxonomy of ai explainabilit...

    work page 2019

  8. [8]

    and Santos, M

    Azevedo, A. and Santos, M. F. (2008). Kdd, semma and crisp-dm: a parallel overview. In Europ. Conf. Data Mining (IADIS) , page 182–185

    work page 2008

Show all 73 references
  1. [9]

    V., Viaene, S., Stepanova, M., Suykens, J., and Vanthienen, J

    Baesens, B., Gestel, T. V., Viaene, S., Stepanova, M., Suykens, J., and Vanthienen, J. (2002). Benchmarking state-of-the-art classification algorithms for credit scoring. JORS , 54(6):627--635

    2002

  2. [10]

    and Crook, J

    Banasik, J. and Crook, J. (2007). Reject inference, augmentation and sample selection. European Journal of Operational Research , 183:1582--1594

    2007

  3. [11]

    and Crook, J

    Bellotti, T. and Crook, J. (2009). Support vector machines for credit scoring and discovery of significant features. Expert Systems with Applications , 2(33):3302--3308

    2009

  4. [12]

    Biecek, P. (2018). Dalex: explainers for complex predictive models. Journal of Machine Learning Research , 19(84):1--5

    2018

  5. [13]

    and Burzykowski, T

    Biecek, P. and Burzykowski, T. (2019). Explanatory model analysis. explore, explain and examine predictive models. online

    2019

  6. [14]

    Bischl, B., K \"u hn, T., and Szepannek, G. (2014). On class imbalance correction for classification algorithms in credit scoring. In Lübbecke, M., Koster, A., P., L., R., M., B., P., and Walther, G., editors, Operations Research Proceedings , pages 37--43

    2014

  7. [15]

    and Christophe, M

    Brown, I. and Christophe, M. (2012). An experimental comparison of classification algorithms for imbalanced credit scoring data sets. Expert Systems with Applications , 3(39):3446--3453

    2012

  8. [16]

    Bücker, M., van Kampen, M., and Krämer, W. (2013). Reject inference in consumer credit scoring with nonignorable missing data. Journal of Banking & Finance , 37(3):1040--1045

    2013

  9. [17]

    An explainable model for credit risk performance

    Chen, C., Lin, K., Rudin, C., Shaposhnik, Y., Wang, S., and Wang, T. An explainable model for credit risk performance. Explainable Machine Learning Challenge documentation

  10. [18]

    and Guestrin, C

    Chen, T. and Guestrin, C. (2016). Xgboost: A scalable tree boosting system. In Proceedings of the 22Nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD '16, pages 785--794, New York, NY, USA. ACM

    2016

  11. [19]

    Cook, D. (2016). Practical Machine Learning with H2O: Powerful, Scalable Techniques for Deep Learning and AI . O'Reilly Media

    2016

  12. [20]

    and Vapnik, V

    Cortes, C. and Vapnik, V. (1995). Support-vector networks. Machine Learning , 20(3):273--297

    1995

  13. [21]

    Crook, J., Edelman, D., and Thomas, L. C. (2007). Recent developments in consumer credit risk assessment. Journal of the Operational Research Society , (183):1447--1465

    2007

  14. [22]

    Dash, S., G\" u nl\" u k, O., and Wei, D. (2018). Boolean decision rules via column generation. In Proceedings of the 32nd International Conference on Neural Information Processing Systems , NIPS’18, page 4660–4670, Red Hook, NY, USA. Curran Associates Inc

    2018

  15. [23]

    Ethics guidelines for trustworthy ai

    EU Expert Group on AI (2019). Ethics guidelines for trustworthy ai. Online

    2019

  16. [24]

    Guidelines on pd estimation, lgd estimation and the treatment of defaulted exposures

    European Banking Authority (2017). Guidelines on pd estimation, lgd estimation and the treatment of defaulted exposures. Online

    2017

  17. [25]

    On artificial intelligence - a european approach to excellence and trust

    European Commission (2020). On artificial intelligence - a european approach to excellence and trust. Online

    2020

  18. [26]

    European Union, G. D. P. R. (2016). Regulation (eu) 2016/679 of the european parliament and of the council

    2016

  19. [27]

    xml challenge

    FICO (2019). xml challenge. Online

    2019

  20. [28]

    Artificial intelligence and machine learning in financial services -- market developments and financial stability implications

    Financial Stability Board (2017). Artificial intelligence and machine learning in financial services -- market developments and financial stability implications. Online

    2017

  21. [29]

    Finlay, S. (2012). Credit Scoring, Response Modelling and Insurance Rating . Palgarve MacMillan

    2012

  22. [30]

    Fisher, A., Rudin, C., and Dominici, F. (2018). Model class reliance: Variable importance measures for any machine learning model class, from the 'rashomon' perspective. Journal of Computational and Graphical Statistics

    2018

  23. [31]

    and Mues, C

    Fitzpatrick, T. and Mues, C. (2016). An empirical comparison of classification algorithms for mortgage default prediction: evidence from a distressed mortgage market. European Journal of Operational Research , 2(249):427--439

    2016

  24. [32]

    Friedman, J., Hastie, T., and Tibshirani, R. (2010). Regularization paths for generalized linear models via coordinate descent. Journal of Statistical Software , 33(1):1--22

    2010

  25. [33]

    Friedman, J. H. (2000). Greedy function approximation: A gradient boosting machine. Annals of Statistics , 29:1189--1232

    2000

  26. [34]

    and Steuer, D

    Garzcarek, U. and Steuer, D. (2019). Approaching Ethical Guidelines for Data Scientists , pages 151--169. Springer International Publishing

    2019

  27. [35]

    and Hall, P

    Gill, N. and Hall, P. (2018). An introduction to machine learning interpretability. O'Reilly Media, Inc

    2018

  28. [36]

    Goldstein, A., Kapelner, A., Bleich, J., and Pitkin, E. (2015). Peeking inside the black box: Visualizing statistical learning with plots of individual conditional expectation. Journal of Computational and Graphical Statistics , 24(1):44--65

    2015

  29. [37]

    Gomez, O., Holter, S., Yuan, J., and Bertini, E. (2020). Vice: Visual counterfactual explanations for machine learning models. In Proceedings of the 25th International Conference on Intelligent User Interfaces , IUI ’20, page 531–535. ACM

    2020

  30. [38]

    right to explanation

    Goodman, B. and Flaxman, S. (2017). European union regulations on algorithmic decision-making and a “right to explanation”. AI Magazine , 38(3):50--57

    2017

  31. [39]

    and Biecek , P

    Gosiewska , A. and Biecek , P. (2019). Do Not Trust Additive Explanations . arXiv e-prints , page arXiv:1903.11420

    2019 arXiv

  32. [40]

    Greenwell, B., Boehmke, B., Cunningham, J., and Developers, G. (2019). gbm: Generalized Boosted Regression Models . R package version 2.1.5

    2019

  33. [41]

    Greenwell, B. M. (2017). pdp: An r package for constructing partial dependence plots. The R Journal , 9(1):421--436

    2017

  34. [42]

    Hand, D. (2009). Measuring classifier performance: a coherent alternative to the area under the roc curve. Machine Learning , 77:103–--123

    2009

  35. [43]

    Harrell, F. (2015). Regression Modeling Strategies: With Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis . Springer Series in Statistics. Springer International Publishing

    2015

  36. [44]

    Fico explainable machine learning challenge

    Holter, S., Gomez, O., and Bertini, E. Fico explainable machine learning challenge. creating visual explanations to black-box machine learning models. Explainable Machine Learning Challenge documentation

  37. [45]

    Jenkins, S., Nori, H., Koch, P., and Caruana, R. (2019). Interpretml

    2019

  38. [46]

    and Loftus, J

    Kusner, M. and Loftus, J. (2020). The long road to fairer algorithms. Nature , 534:34--36

    2020

  39. [47]

    Lessmann, S., Baesens, B., Seow, H.-V., and Thomas, L. (2015). Benchmarking state-of-the-art classification algorithms for credit scoring: An update of research. European Journal of Operational Research , 247(1):124--136

    2015

  40. [48]

    and Wiener, M

    Liaw, A. and Wiener, M. (2002). Classification and regression by randomforest. R News , 2(3):18--22

    2002

  41. [49]

    Louzada, F., Ara, A., and Fernandes, G. (2016). Classification methods applied to credit scoring: A systematic review and overall comparison. Surveys in OR and Management Science , 21(2):117--134

    2016

  42. [50]

    Luebke, K., Gehrke, M., Horst, J., and Szepannek, G. (2020). Why we should teach causal inference: Examples in linear regression with simulated data. Journal of Statistics Education

    2020

  43. [51]

    Lundberg, S. M. and Lee, S.-I. (2017). A unified approach to interpreting model predictions. In Guyon, I., Luxburg, U. V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., and Garnett, R., editors, Advances in Neural Information Processing Systems 30 , pages 4765--4774. ...

    2017

  44. [52]

    McGough, M. (2018). How bad is Sacramento’s air, exactly? Google results appear at odds with reality, some say . https://www.sacbee.com/news/california/fires/article216227775.html. Accessed: 2019-10-12

    2018

  45. [53]

    Molnar, C. (2019). Interpretable Machine Learning . https://christophm.github.io/interpretable-ml-book/

    2019

  46. [54]

    Molnar, C., Bischl, B., and Casalicchio, G. (2018). iml: An R package for Interpretable Machine Learning . Journal of Open Source Software , 3(26):786

    2018

  47. [55]

    Montavon, G., Samek, W., and Müller, K.-R. (2018). Methods for interpreting and understanding deep neural networks. Digital Signal Processing , 73:1 -- 15

    2018

  48. [56]

    O'Neil, C. (2016). Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy . Crown Publishing Group, New York, NY, USA

    2016

  49. [57]

    Płoński, P. (2019). mljar-supervised: The Automated Machine Learning - the new standard in ML. Machine Learning for Humans

    2019

  50. [58]

    why should I trust you?

    Ribeiro, M. T., Singh, S., and Guestrin, C. (2016). "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, San Francisco, CA, USA, August 13-17, 2016 , page...

    2016

  51. [59]

    Robin, X., Turck, N., Hainard, A., Tiberti, N., Lisacek, F., Sanchez, J., and Müller, M. (2011). proc: an open-source package for r and s+ to analyze and compare roc curves. BMC Bioinformatics , 12

    2011

  52. [60]

    Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence , 1(5):206--215

    2019

  53. [61]

    Scallan, G. (2011). Class(ic) scorecards – selecting attributes in logistic regression. In Credit Scoring and Credit Control XIII

    2011

  54. [62]

    Schölkopf, B. (2019). Causality for machine learning

    2019

  55. [63]

    and Flach, P

    Sokol, K. and Flach, P. (2020). One Explanation Does Not Fit All . KI - Kunstliche Intelligenz

    2020

  56. [64]

    Szepannek, G. (2017a). A framework for scorecard modelling using r. In Credit Scoring and Credit Control XV

    2017

  57. [65]

    Szepannek, G. (2017b). On the practical relevance of modern machine learning algorithms for credit scoring applications. WIAS Report Series , 29:88--96

    2017

  58. [66]

    Szepannek, G. (2019). How much can we see? A note on quantifying explainability of machine learning models. arxiv

    2019

  59. [67]

    and Aschenbruck, R

    Szepannek, G. and Aschenbruck, R. (2019). Predicting ebay prices: Selecting and interpreting machine learning models – results of the ag dank 2018 data science competition. Archives of Data Science A (accepted)

    2019

  60. [68]

    C., Crook, J

    Thomas, L. C., Crook, J. N., and Edelman, D. B. (2019). Credit Scoring and its Applications . SIAM, second edition

    2019

  61. [69]

    and Martens, D

    Tobback, E. and Martens, D. (2019). Retail credit scoring using fine-grained payment data. Journal of the Royal Statistical Society: Series A (Statistics in Society) , 182(4):1227--1246

    2019

  62. [70]

    Verbraken, T., Bravo, C., Richard, W., and Baesens, B. (2014). Development and application of consumer credit scoring models using profit-based classification measures. European Journal of Operational Research , 238(2):505--513

    2014

  63. [71]

    Wexler, R. (2017). When a Computer Program Keeps You in Jail . https://www.nytimes.com/2017/06/13/opinion/how-computers-are-harming-criminal-justice.html. Accessed: 2019-10-12

    2017

  64. [72]

    Wright, M. N. and Ziegler, A. (2017). ranger : A fast implementation of random forests for high dimensional data in C++ and R . Journal of Statistical Software , 77(1):1--17

    2017

  65. [73]

    and Hastie, T

    Zhao, Q. and Hastie, T. (2019). Causal interpretations of black-box models. Journal of Business & Economic Statistics

    2019

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.