Pith. sign in

REVIEW 3 major objections 6 minor 44 references

Evaluating the stability of model explanations in instance-dependent cost-sensitive credit scoring

T0 review · 3 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Cost-sensitive credit models win on savings but lose on explanation stability.

desk verdict A useful empirical study of explanation stability for cost-sensitive credit models; the main trade-off claim is credible but rests on a stability metric with a known signed-value weakness. read the letter →

arxiv 2509.01409 v1 pith:YMIERXIQ submitted 2025-09-01 cs.LG

Matteo Ballegeer , Matthias Bogaert , Dries F. Benoit This is my paper · ORCID
classification cs.LG
keywords cost-sensitivelearningcreditscoringexplanationstabilitySHAPLIMEclassimbalanceAverageExpectedCostinstance-dependentcosts
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

This paper tries to establish that instance-dependent cost-sensitive (IDCS) classifiers—models that tailor misclassification penalties to each loan's individual cost—improve cost-efficiency but produce systematically less stable explanations than traditional cost-insensitive models. Using SHAP and LIME, the two most widely used explanation tools, the authors compare eight classifiers across four public credit datasets and measure how feature importances vary across repeated training runs. They find that IDCS models yield higher dispersion in feature-importance values and rankings, and that this instability grows sharply as class imbalance increases. If true, this reveals a concrete trade-off between profit-driven model design and regulatory demands for transparent, trustworthy explanations.

What carries the argument

The load-bearing object is the instance-dependent cost matrix—a false negative costs the loan amount times a loss-given-default parameter, while a false positive costs lost revenue plus the expected cost of lending to an alternative borrower—folded into an Average Expected Cost (AEC) loss that IDCS classifiers directly minimize. The stability comparison is carried by two metrics: the Coefficient of Variation (CoV), the average across features of the standard deviation divided by the mean of importance values over repetitions, and Sequential Rank Agreement (SRA), the pooled variance of feature ranks across repeated importance lists. The newly introduced relAEC metric normalizes AEC against a

What would settle it

Recompute the stability comparison using absolute importance values or a signed-mean-adjusted dispersion measure, such as median absolute deviation or variance of ranks without the CoV ratio, on the same 25-iteration resamples; if IDCS and traditional models show similar dispersion under that metric, the claimed stability gap is an artifact of the CoV formula, while if the gap persists the claim survives.

Watch

Extended reading notes

Core claim

The paper's central claim is that IDCS classifiers achieve better cost-sensitive performance but create significantly less stable local explanations. Across four datasets and four model families (boosted trees, logistic regression, random forest, and neural networks) paired with their IDCS variants, SHAP and LIME feature importances from IDCS models show higher Coefficient of Variation (CoV) values and higher Sequential Rank Agreement (SRA) values than their cost-insensitive twins, indicating more variability across 25 resampling iterations. This instability becomes more pronounced as the training default rate is resampled toward extreme imbalance (e.g., 1% and 3%). The paper also introduces

Load-bearing premise

The headline comparison treats the coefficient of variation (standard deviation divided by mean) of feature importances as a valid stability measure, even though SHAP values and LIME coefficients can be negative or near zero, where a large ratio can reflect a small denominator rather than genuine instability.

Editorial extensions

If this is right

  • Banks adopting IDCS credit models should expect explanation instability as a systematic side effect, not a tuning artifact.
  • Under imbalanced loan portfolios, explanations for why a loan was denied may vary substantially between otherwise similar applications, complicating customer-facing justification.
  • Model selection for credit scoring should weigh cost-efficiency against explanation stability; the financially best model may not be the most explainable one.
  • Evaluation protocols for explanation methods in credit should include class-imbalance sweeps, because stability degrades nonlinearly at extreme default rates.
  • Repeated-retraining stability checks, like the 25-iteration resampling used in this paper, belong in model risk management for regulated credit models.

Reading between the lines

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

  • If the CoV metric is sensitive to near-zero and sign-changing SHAP values, part of the observed stability gap may reflect IDCS models concentrating importance in small signed values; a follow-up using absolute values or rank-based dispersion would separate genuine instability from a ratio artifact.
  • The paper's proposed mechanism—costs entangle features, labels, and loan amounts—suggests that datasets with narrower false-negative-to-false-positive cost ratio distributions might show smaller stability gaps; this is testable with synthetic data.
  • The relAEC normalization could generalize beyond credit scoring to any instance-dependent cost setting, though its usefulness depends on the prior default rate being a sensible 'no model' baseline.
  • Class-imbalance-driven instability in IDCS explanations suggests that cost-aware stabilization techniques, such as smoothing or constrained optimization on feature attributions, could be developed to preserve cost efficiency without sacrificing explainability.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper investigates whether instance-dependent cost-sensitive (IDCS) classifiers in credit scoring yield less stable post-hoc explanations than standard classifiers. The authors benchmark eight models (XGBoost, logistic regression, random forest, neural network, and their IDCS variants) on four public credit-scoring datasets, measuring performance with AUC, AP, Brier, savings, and a newly proposed relAEC metric. Explanation stability is assessed for SHAP and LIME via the coefficient of variation (CoV) and sequential rank agreement (SRA) under controlled resampling to training default rates between 1% and 30%. The central claim is that IDCS models improve cost-efficiency but produce significantly less stable explanations, and that the stability gap grows as class imbalance increases.

Significance. If the result holds, it identifies a concrete, regulation-relevant trade-off between profit-driven loss functions and explanation stability, which is especially important in credit scoring under GDPR and the EU AI Act. The study is carefully designed on the whole: nested cross-validation for performance, a fixed test set, 25 resampling iterations per imbalance level, two complementary stability metrics, and four model families. The SRA results provide independent support for the main stability claim in most settings, and the relAEC metric is a useful normalization. The main caveat is that the value-based CoV metric, as defined in Eq. (6), is sensitive to near-zero signed means; the manuscript does not address this issue, and the interaction claim that instability increases 'particularly' for IDCS under class imbalance is not backed by a statistical test.

major comments (3)
  1. [Section 3.6, Eq. (6), and Section 3.5] The manuscript does not state whether the CoV is computed on raw SHAP values/LIME coefficients or on their absolute values. Section 3.5 says that feature importances are derived from SHAP values and LIME coefficients, with 'higher absolute values indicating greater influence,' but Eq. (6) defines the CoV as sigma_ip/mu_ip. SHAP values and LIME coefficients are signed; if raw values are used, mu_ip near zero can inflate CoV arbitrarily, and a model with small, sign-changing attributions would appear unstable even when its absolute importance is consistent. Since the CoV gap between IDCS and traditional models is a headline result, please state explicitly which values were used. If raw values were used, repeat the analysis with absolute importances and report whether the gap persists; also state how zero or near-zero means were handled.
  2. [Section 4.2 (Figs. 4–7) and Section 6] The abstract and conclusions state that IDCS explanations become less stable 'particularly as class imbalance increases,' but the paper reports no statistical test for an interaction between model type and imbalance level. The figures show higher CoV/SRA for IDCS at most levels, but the stronger-imbalance-effect claim requires e.g. a regression of stability on imbalance and model type with an interaction term, or a permutation test. This is especially important because SHAP SRA is explicitly non-monotonic for VUB and SGCS, and LIME SRA is not significantly higher for SGCS. The abstract should qualify the claim by these exceptions or report an appropriate interaction test.
  3. [Section 4.2.1] The text states that a two-sample Kolmogorov-Smirnov test 'confirms that the CoV distribution of IDCS classifiers is significantly higher than that of traditional classifiers.' The standard two-sample KS test is two-sided and tests for any distributional difference, not specifically for stochastic dominance. Please report a one-sided test, or an effect size and direction, to support the directional claim of 'significantly higher.'
minor comments (6)
  1. [Section 2] Typo: 'analaysis' should be 'analysis'.
  2. [Figure 8 caption] Typo: 'defualter' should be 'defaulter'.
  3. [Section 4.1] Typo: 'optimizning' should be 'optimizing'.
  4. [Figures 4–7] In the provided version, axis labels and panel labels appear as garbled character sequences (e.g., '/uni00000014...'), making the figures hard to read. Please ensure the final PDF rendering is clean.
  5. [Algorithm 2] The line 'Set tuning_round To T rue' should read 'Set tuning_round To True'.
  6. [Section 3.8] Please state explicitly how ties in feature importance rankings are handled when computing SRA, and whether the 300 test observations are held fixed across all resampling iterations (as opposed to resampled per iteration).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central trade-off claim is an empirical measurement with externally defined stability metrics.

full rationale

The paper's central claim—that IDCS classifiers improve cost-efficiency but yield less stable SHAP/LIME explanations, especially under class imbalance—is an empirical comparison rather than a derivation from fitted constants. The stability metrics are predefined and externally sourced: CoV is the standard coefficient of variation (Eq. 6), and SRA follows Ekstrøm et al. (2018) (Eqs. 7–8). The IDCS/traditional gap is read off measured distributions in Figures 4–7, not constructed by the metric definitions. The relAEC metric (Eq. 5) is a dimensionless normalization of AEC against a prior-probability baseline; it is an evaluation score, and the stability conclusion does not depend on it. Models and cost matrices are imported from prior work (Bahnsen et al., Höppner et al., Vanderschueren et al.), which is external implementation reuse, not a self-justifying uniqueness theorem. The few self-citations (e.g., Janssens et al. 2022, 2023) are background context and do not carry the trade-off claim. The potential concern about CoV being applied to signed SHAP/LIME values is a measurement-validity issue, not a circular reduction: no equation in the paper equates the stability gap to an input parameter or fitted target. Thus the derivation chain is self-contained with respect to the paper's stated inputs.

Assumptions & free parameters 3 free parameters · 6 assumptions · 1 invented entities

The central claim depends on the choice of cost matrix, the fixed prior rates during resampling, the choice of SHAP/LIME implementations, and the validity of the two stability metrics. None of these is fitted to the stability outcome, but all are domain assumptions that could change the measured gap. relAEC is the only newly introduced construct, and it has no independent benchmark.

free parameters (3)
  • LGD = 0.75
    Fixed Loss Given Default constant from Global Credit Data (2023) used in every instance cost matrix. It is chosen by the authors and affects all IDCS training, but it is not fitted to the outcome.
  • Average alternative-customer profit r and average loan amount A = derived from training-set cost data, not reported in the preprint
    Used in Eq. 1 to define the false-positive cost. The paper states A is derived from the training set but does not report the resulting value or the estimation procedure for r.
  • SRA depth = 10
    The ranked feature lists are compared only for features appearing in the top 10 of at least one list. This choice affects the measured rank stability and can weaken or strengthen the observed gap.
assumptions (6)
  • domain assumption The Bahnsen et al. (2014) cost matrix with LGD = 0.75 and an alternative-customer term is the correct way to set instance-dependent costs in credit scoring.
    Section 3.3 and Table 3. All IDCS models are trained under this cost model; a different cost matrix could change the stability result.
  • domain assumption Fixed prior rates pi_0 and pi_1 during resampling keep the cost matrix comparable across imbalance levels.
    Section 3.3, final paragraph. This prevents the alternative-customer cost from varying with resampling, but it is a modeling decision that shapes the stability comparison.
  • domain assumption SHAP PermutationExplainer and LIME with an Euclidean-distance kernel produce feature importances suitable for stability analysis.
    Section 3.5. The entire stability comparison depends on these explainer implementations and their default randomness.
  • domain assumption CoV and SRA with depth 10 are valid and sufficient operationalizations of explanation stability.
    Section 3.6. If CoV is invalid for signed importances, part of the central evidence is compromised.
  • domain assumption The four selected public datasets are representative of credit scoring applications for the purpose of the trade-off claim.
    Section 3.1. Generalization to other cost distributions and data sizes is assumed, and Section 5 notes dataset-specific exceptions.
  • standard math Standard supervised learning assumptions hold: features X, binary outcome Y, and s(x) = E(Y|x) are correctly estimated by the classifiers.
    Eq. 2 and the model definitions. The stability analysis interprets SHAP and LIME as explanations of this estimated probability.
invented entities (1)
  • relAEC metric
    purpose: Dimensionless normalization of Average Expected Cost for cross-dataset comparison.
    Introduced in Section 3.4, Eq. 5. It is a new summary statistic; the paper provides no external validation that its ranking is more meaningful than savings or raw AEC.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evaluating the stability of model explanations in instance-dependent cost-sensitive credit scoring." pith.science (2026). https://pith.science/paper/YMIERXIQ

@misc{pith2026250901409,
  author       = {Pith},
  title        = {Pith review of: Evaluating the stability of model explanations in instance-dependent cost-sensitive credit scoring},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YMIERXIQ}},
  note         = {Machine review of arXiv:2509.01409}
}
read the original abstract

Instance-dependent cost-sensitive (IDCS) classifiers offer a promising approach to improving cost-efficiency in credit scoring by tailoring loss functions to instance-specific costs. However, the impact of such loss functions on the stability of model explanations remains unexplored in literature, despite increasing regulatory demands for transparency. This study addresses this gap by evaluating the stability of Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) when applied to IDCS models. Using four publicly available credit scoring datasets, we first assess the discriminatory power and cost-efficiency of IDCS classifiers, introducing a novel metric to enhance cross-dataset comparability. We then investigate the stability of SHAP and LIME feature importance rankings under varying degrees of class imbalance through controlled resampling. Our results reveal that while IDCS classifiers improve cost-efficiency, they produce significantly less stable explanations compared to traditional models, particularly as class imbalance increases, highlighting a critical trade-off between cost optimization and interpretability in credit scoring. Amid increasing regulatory scrutiny on explainability, this research underscores the pressing need to address stability issues in IDCS classifiers to ensure that their cost advantages are not undermined by unstable or untrustworthy explanations.

Figures

Figures reproduced from arXiv: 2509.01409 by the authors.

Figure 1
Figure 1. Proposed experimental setup set, evaluating stability between the feature importance lists via the coefficient of variation (CoV) and Sequential Rank Agreement (SRA) (Ekstrøm et al., 2018). Each step of our proposed methodology is presented in the following sections. 3.1 Data The four publicly available credit scoring datasets are selected because of their popularity in research (Vanderschueren et al., 2022b; Devos … view at source ↗
Figure 2
Figure 2. Example SHAP CoV computation across 3 iterations. For each feature, we first compute the CoV value across iterations, after which we average the CoV of all the features to get an estimate per instance [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Baseline savings example with one actual defaulter and 2 non-defaulters [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Distribution of SHAP CoV values At all imbalance levels, SHAP explanations from IDCS models consistently exhibit higher CoV values than those from cost-insensitive models. This indicates greater variability in SHAP feature importance values across iterations, resulting…
Figure 5
Figure 5. Figure 5: Distribution of LIME CoV values A similar trend is observed for LIME, as shown in [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: Distribution of SHAP SRA values The findings for SHAP and LIME SRA are consistent with those for SHAP CoV, as IDCS classifiers generally demonstrate significantly higher SRA values across all levels of imbalance, with instability increasing as class imbalance grows, ex…
Figure 7
Figure 7. Figure 7: Distribution of LIME SRA values observed for traditional classifiers by Chen et al. (2024), it is more pronounced for IDCS classifiers. This may be due to the combined effect of a scarcity of positive samples and noisier feature-label relationships, which provide insuf…
Figure 8
Figure 8. Figure 8: Cost ratios across datasets. The x-axis represents the false negative to false positive F N F P cost ratio for every actual defualter, derived from [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 25 canonical work pages

  1. [8]

    European Journal of Operational Research 312, 357–372

    Interpretable machine learning for imbalanced credit scoring datasets. European Journal of Operational Research 312, 357–372. doi:doi:10.1016/j.ejor.2023.06.036. Credit Fusion, W.C.,

    doi:10.1016/j.ejor.2023.06.036 2023

  2. [10]

    European Journal of Operational Research 317, 249–272

    Explainable ai for operational research: A defining framework, methods, applications, and a research agenda. European Journal of Operational Research 317, 249–272. doi:doi:10.1016/j.ejor.2023.09.026. De V os, S., en, T., Verdonck, T., Verbeke, W.,

    doi:10.1016/j.ejor.2023.09.026 2023

  3. [12]

    Profit maximizing logistic regression modeling for credit scoring, in: 2018 IEEE Data Science Workshop (DSW), IEEE. pp. 125–129. doi:doi:10.1109/DSW.2018.8439113. 19 ACCEPTED MANUSCRIPT VERSION, PUBLISHED IN THE EUROPEAN JOURNAL OF OPERATIONAL RESEARCH Doumpos, M., Zopounidis, C., Gounopoulos, D., Platanakis, E., Zhang, W.,

    work page arXiv 2018

  4. [13]

    European Journal of Operational Research 306, 1–16

    Operational research and artificial intelligence methods in banking. European Journal of Operational Research 306, 1–16. doi:doi:10.1016/j.ejor.2022.04.027. Dumitrescu, E., Hué, S., Hurlin, C., Tokpavi, S.,

    work page doi:10.1016/j.ejor.2022.04.027 2022

  5. [14]

    European Journal of Operational Research 297, 1178–1192

    Machine learning for credit scoring: Improving logistic regression with non-linear decision-tree effects. European Journal of Operational Research 297, 1178–1192. doi:doi:10.1016/j.ejor.2021.06.053. Ekstrøm, C.T., Gerds, T.A., Jensen, A.K.,

    work page doi:10.1016/j.ejor.2021.06.053 2021

  6. [15]

    Biostatistics 20, 582–598

    Sequential rank agreement methods for comparison of ranked lists. Biostatistics 20, 582–598. doi:doi:10.1093/biostatistics/kxy017. Elkan, C.,

    work page doi:10.1093/biostatistics/kxy017

  7. [17]

    (Eds.), Contributions to the Theory of Games, V olume II

    a value for n-person games, in: Kuhn, H.W., Tucker, A.W. (Eds.), Contributions to the Theory of Games, V olume II. Princeton University Press, Princeton, pp. 307–318. doi:doi:doi:10.1515/9781400881970-018. Slack, D., Hilgard, S., Jia, E., Singh, S., Lakkaraju, H.,

    doi:10.1515/9781400881970-018

  8. [18]

    Report EBA/REP/2023/28

    Machine Learning for IRB Models: Follow-up Report from the Consultation on the Discussion Paper on Machine Learning for IRB Models. Report EBA/REP/2023/28. European Parliament,

    work page 2023

Show all 44 references
  1. [19]

    URL: http://data.europa.eu/eli/reg/2024/1689/oj

    Regulation (eu) 2024/1689 of the european parliament and of the council of 13 june 2024 laying down harmonised rules on artificial intelligence and amending various regulations and directives (artificial intelligence act). URL: http://data.europa.eu/eli/reg/2024/1689/oj. Garcí...

    2024

  2. [20]

    Technical Report

    Corporates Annual observed recovery rate trends. Technical Report. Global Credit Data. URL: https://globalcreditdata.org/gcd_library/pd-report-large-corporates-2023/ . Goldwasser, J., Hooker, G.,

    2023

  3. [21]

    arXiv preprint arXiv:2401.15800 doi:doi:10.48550/arXiv.2401.15800

    Provably stable feature rankings with shap and lime. arXiv preprint arXiv:2401.15800 doi:doi:10.48550/arXiv.2401.15800. Goodman, B., Flaxman, S.,

    doi arXiv

  4. [24]

    SSRN Electronic Journal doi:doi:10.2139/ssrn.4406486

    Supervised learning for integrated forecasting and inventory control. SSRN Electronic Journal doi:doi:10.2139/ssrn.4406486. Höppner, S., Baesens, B., Verbeke, W., Verdonck, T.,

    doi

  5. [25]

    European Journal of Operational Research 297, 291–300

    Instance-dependent cost-sensitive learning for detecting transfer fraud. European Journal of Operational Research 297, 291–300. doi:doi:10.1016/j.ejor.2021.05.028. Janssens, B., Bogaert, M., Bagué, A., Van den Poel, D.,

    2021 doi

  6. [26]

    Annals of Operations Research , 1–27doi:doi:10.1007/s10479-022-04631-5

    B2boost: Instance-dependent profit-driven modelling of b2b churn. Annals of Operations Research , 1–27doi:doi:10.1007/s10479-022-04631-5. Janssens, B., Schetgen, L., Bogaert, M., Meire, M., Van den Poel, D.,

    doi

  7. [27]

    European Journal of Operational Research doi:doi:10.1016/j.ejor.2023.06.024

    360 degrees rumor detection: When expla- nations got some explaining to do. European Journal of Operational Research doi:doi:10.1016/j.ejor.2023.06.024. Kamalov, F., Atiya, A.F., Elreedy, D.,

    2023 doi

  8. [28]

    arXiv preprint arXiv:2207.04631 doi:doi:10.48550/arXiv.2207.04631

    Partial resampling of imbalanced data. arXiv preprint arXiv:2207.04631 doi:doi:10.48550/arXiv.2207.04631. Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., Liu, T.Y .,

    doi arXiv

  9. [30]

    neurips.cc/paper_files/paper/2017/file/6449f44a102fde848669bdd9eb6b76fa-Paper.pdf

    URL: https://proceedings. neurips.cc/paper_files/paper/2017/file/6449f44a102fde848669bdd9eb6b76fa-Paper.pdf. 20 ACCEPTED MANUSCRIPT VERSION, PUBLISHED IN THE EUROPEAN JOURNAL OF OPERATIONAL RESEARCH Lessmann, S., Baesens, B., Seow, H.V ., Thomas, L.C.,

    2017

  10. [31]

    European Journal of Operational Research 247, 124–136

    Benchmarking state-of-the-art classification algo- rithms for credit scoring: An update of research. European Journal of Operational Research 247, 124–136. doi:doi:10.1016/j.ejor.2015.05.030. Li, Z., Zhang, J., Yao, X., Kou, G.,

    2015 doi

  11. [32]

    Knowledge-Based Systems 221, 106963

    How to identify early defaults in online lending: A cost-sensitive multi-layer learning framework. Knowledge-Based Systems 221, 106963. doi:doi:10.1016/j.knosys.2021.106963. Lundberg, S.M., Lee, S.I.,

    2021

  12. [33]

    (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc

    A unified approach to interpreting model predictions, in: Guyon, I., Luxburg, U.V ., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., Garnett, R. (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc.. pp. 4768–4777. URL: https://proceedings.n...

    2017

  13. [34]

    Annals of Operations Research , 1–26doi:doi:10.1007/s10479-025-06514- x

    A novel financial performance metric to minimize misclassification costs in model selection. Annals of Operations Research , 1–26doi:doi:10.1007/s10479-025-06514- x. Petrides, G., Moldovan, D., Coenen, L., Guns, T., Verbeke, W.,

    doi

  14. [36]

    why should I trust you?

    "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, pp. 1135–1144. doi:doi:10.1145/2939672.2939778. Saito, T...

    2016

  15. [37]

    PloS one 10, e0118432

    The precision-recall plot is more informative than the roc plot when evaluating binary classifiers on imbalanced datasets. PloS one 10, e0118432. doi:doi:10.1371/journal.pone.0118432. Shapley, L.S.,

    doi

  16. [39]

    Knowledge and Information Systems 41, 647–665

    Explaining prediction models and individual predictions with feature contributions. Knowledge and Information Systems 41, 647–665. doi:doi:10.1007/s10115-013-0679-x. Szepannek, G., Holt, B.H.v.,

    doi

  17. [40]

    Behaviormetrika 51, 411–423

    Can’t see the forest for the trees: Analyzing groves to explain random forests. Behaviormetrika 51, 411–423. doi:doi:10.1007/s41237-023-00205-2. UCI Machine Learning Repository,

    doi

  18. [42]

    European Journal of Operational Research 238, 505–513

    Development and application of consumer credit scoring models using profit-based classification measures. European Journal of Operational Research 238, 505–513. doi:doi:10.1016/j.ejor.2014.04.001. Visani, G., Bagli, E., Chesani, F., Poluzzi, A., Capuzzo, D.,

    2014 doi

  19. [43]

    Journal of the Operational Research Society 73, 91–101

    Statistical stability indices for LIME: Obtaining reliable explanations for machine learning models. Journal of the Operational Research Society 73, 91–101. doi:doi:10.1080/01605682.2020.1865846. Wu, S., Gao, X., Zhou, W.,

    2020

  20. [44]

    Information Sciences 586, 74–98

    Cosle: Cost sensitive loan evaluation for p2p lending. Information Sciences 586, 74–98. doi:doi:10.1016/j.ins.2021.11.055. 21 ACCEPTED MANUSCRIPT VERSION, PUBLISHED IN THE EUROPEAN JOURNAL OF OPERATIONAL RESEARCH Zelenkov, Y .,

    2021 doi

  21. [45]

    Expert Systems with Applications 135, 71–82

    Example-dependent cost-sensitive adaptive boosting. Expert Systems with Applications 135, 71–82. doi:doi:10.1016/j.eswa.2019.06.009. Zhou, Z., Hooker, G., Wang, F.,

    2019 doi

  22. [46]

    2429–2438

    S-lime: Stabilized-lime for model explanation, in: Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, pp. 2429–2438. doi:doi:10.1145/3447548.3467274. 22

  23. [2001]

    Proceedings of the Seventeenth International Conference on Artificial Intelligence: 4-10 August 2001; Seattle

    The foundations of cost-sensitive learning. Proceedings of the Seventeenth International Conference on Artificial Intelligence: 4-10 August 2001; Seattle

    2001

  24. [2004]

    Computational intelligence 20, 18–36

    A multiple resampling method for learning from imbalanced data sets. Computational intelligence 20, 18–36. doi:doi:10.1111/j.0824-7935.2004.t01-1-00228.x. European Banking Authority,

    2004

  25. [2006]

    The relationship between precision-recall and roc curves, in: Proceedings of the 23rd international conference on Machine learning, pp. 233–240. doi:doi:10.1145/1143844.1143874. De Bock, K.W., Coussement, K., Caigny, A.D., Słowi ´nski, R., Baesens, B., Boute, R.N., Choi, T.M.,...

  26. [2014]

    Example-dependent cost-sensitive logistic regression for credit scoring, in: 2014 13th International conference on machine learning and applications, IEEE. pp. 263–269. doi:doi:10.1109/ICMLA.2014.48. Bahnsen, A.C., Aouada, D., Ottersten, B., 2015a. Ensemble of example-dependen...

    2014 doi arXiv

  27. [2015]

    John Wiley & Sons

    Fraud analytics using descriptive, predictive, and social network techniques: a guide to data science for fraud detection. John Wiley & Sons. doi:doi:10.1002/9781119146841. Bahnsen, A.C., Aouada, D., Ottersten, B.,

    doi

  28. [2017]

    right to explanation

    European union regulations on algorithmic decision-making and a “right to explanation”. AI magazine 38, 50–57. doi:doi:10.1609/aimag.v38i3.2741. Gunnarsson, B.R., vanden Broucke, S., Baesens, B., Óskarsdóttir, M., Lemahieu, W.,

    doi

  29. [2018]

    doi:doi:10.48550/arXiv.1806.08049, arXiv:1806.08049

    On the robustness of interpretability methods. doi:doi:10.48550/arXiv.1806.08049, arXiv:1806.08049. Ariza-Garzón, M.J., Arroyo, J., Segovia-Vargas, M.J., Caparrini, A.,

    doi arXiv

  30. [2019]

    doi:doi:10.24432/C5X89F

    South German Credit. doi:doi:10.24432/C5X89F. Vanderschueren, T., Verdonck, T., Baesens, B., Verbeke, W., 2022a. Instance-dependent cost-sensitive learning: do we really need it?, in: 55th Hawaii International Conference on System Sciences (HICSS-55), pp. 1–9. doi:doi:https://...

    2022 doi

  31. [2020]

    Fooling lime and shap: Adversarial attacks on post hoc explanation methods, in: Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, pp. 180–186. doi:doi:10.1145/3375627.3375830. Štrumbelj, E., Kononenko, I.,

  32. [2021]

    doi:doi:10.1016/j.ejor.2021.03.006

    Deep learning for credit scoring: Do or don’t? European Journal of Operational Research 295, 292–305. doi:doi:10.1016/j.ejor.2021.03.006. Haar, J., Wellens, A., Boute, R., Basten, R.,

    2021 doi

  33. [2022]

    Journal of the Operational Research Society 73, 70–90

    Transparency, auditability, and explainabil- ity of machine learning models in credit scoring. Journal of the Operational Research Society 73, 70–90. doi:doi:10.1080/01605682.2021.1922098. C-Rella, J., Claeskens, G., Cao, R., Vilar, J.M.,

    2021

  34. [2023]

    Advances in Data Analysis and Classification , 1–23doi:doi:10.1007/s11634-022-00533-3

    Robust instance-dependent cost-sensitive classification. Advances in Data Analysis and Classification , 1–23doi:doi:10.1007/s11634-022-00533-3. Devos, A., Dhondt, J., Stripling, E., Baesens, B., vanden Broucke, S., Sukhatme, G.,

    doi

  35. [2024]

    Electronic Commerce Research and Applications 67, 101428

    Profit-sensitive machine learning classifica- tion with explanations in credit risk: The case of small businesses in peer-to-peer lending. Electronic Commerce Research and Applications 67, 101428. doi:doi:10.1016/j.elerap.2024.101428. Baesens, B., Roesch, D., Scheule, H.,

    2024

  36. [2025]

    Neuro- computing 615, 128875

    Instance-dependent cost-sensitive parametric learning. Neuro- computing 615, 128875. doi:doi:10.1016/j.neucom.2024.128875. Chen, T., Guestrin, C.,

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.