Pith. sign in

REVIEW 4 major objections 4 minor 66 references

Software Engineering Principles for Fairer Systems: Experiments with GroupCART

T0 review · 4 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read GroupCART builds fairness into decision trees and keeps accuracy nearly intact.

desk verdict Plausible incremental ensemble method, but the headline 'optimal trade-off' leans on an inverted disparate-impact direction and an under-specified early-stopping rule. read the letter →

arxiv 2504.12587 v1 pith:DHNTZNI4 submitted 2025-04-17 cs.LG cs.SE

Kewen Peng , Hao Zhuo , Yicheng Yang , Tim Menzies This is my paper · ORCID
classification cs.LGcs.SE
keywords GroupCARTalgorithmicbiasfairness-awaredecisiontreesensemblelearningmulti-objectiveoptimizationParetofrontierfairnessindividual
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 sets out to show that algorithmic bias in decision-tree models can be reduced during model construction rather than fixed afterward. It introduces GroupCART, an ensemble of fairness-aware decision trees in which each tree is built with a different weight on information gain from protected attributes. The candidate trees are filtered to the Pareto frontier of the performance-fairness trade-off and combined by majority voting. The paper reports that on seven fairness datasets this yields group-fairness metrics comparable to or better than four existing mitigation methods, with flip rates usually zero and only a small loss in accuracy or F1.

What carries the argument

The central mechanism is the fairness-aware decision tree (FDT), which replaces the usual information-gain split criterion with a weighted sum of $IG_C$, the information gain on the class attribute, and $IG_S$, the information gain on the protected attribute. GroupCART instantiates $N$ such trees with weight ratios $i/N$ versus $1-i/N$, evaluates all candidates on validation data, and applies non-dominated sorting under continuous domination, a softened dominance rule that gives each candidate a scalar loss, to retain the Pareto frontier. Majority voting over those frontier models produces the final prediction. The weight ratio is the hyperparameter that moves the ensemble along the fairness-performance spectrum, and the voting step is what turns several trade-off-optimal configurations into one model.

What would settle it

Re-run GroupCART on the Adult dataset with protected attribute 'sex', computing the flip rate used for early stopping only on the validation split, and check whether the ensemble still stops at 20 members and whether the held-out test flip rate remains zero. If the test FR is nonzero or the ensemble sizes change, the reported perfect individual-fairness results are leakage artifacts.

Watch

Extended reading notes

Core claim

The paper's central claim is that optimizing tree splits for two objectives at once--decreasing entropy in the target label and increasing entropy in protected attributes--and then ensembling the resulting Pareto-optimal configurations produces models that sit on a better performance-fairness frontier than plain CART and than prior mitigation methods. GroupCART builds $N$ candidate trees with distinct $IG_C / IG_S$ weight ratios, keeps the non-dominated ones under continuous domination, and averages their predictions. In the reported experiments it matches or beats EGR, Fax-AI, MAAT, and FairMask on most datasets, reduces flip rate to zero in nearly all settings, and gives best distance-to-heaven scores in several cases while keeping performance close to the unconstrained baseline. The authors conclude from this that algorithmic bias within decision tree models can be mitigated through multi-task fairness-aware learning.

Load-bearing premise

The results assume that the early-stopping rule that fixes ensemble size at 5, 10, and 20--by stopping when flip rate reaches zero--uses only validation data; the paper never states which data split drives that rule, so the reported test flip rates may not be independent if the test set was involved.

Editorial extensions

If this is right

  • Fairness can be treated as a configurable property of tree-based models, so bias mitigation becomes a search over weight ratios rather than a separate repair step.
  • Ensembles with more than two members can improve the fairness-performance trade-off, suggesting that two-model ensembles leave room for improvement.
  • Users can dial a continuous weight between performance and fairness, producing a family of deployable models rather than a single compromise.
  • Multiple protected attributes can be handled at once, so a model can be tuned for fairness on sex and race simultaneously.

Reading between the lines

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

  • If the early-stopping rule is validated as leakage-free, the reported flip rate of zero for nearly all settings would indicate that ensemble voting alone can enforce individual fairness without per-instance constraints; this could be tested by recording FR as a function of ensemble size on held-out data.
  • The same weight-ratio construction should transfer to random forests and AdaBoost-style ensembles; the paper suggests it but does not evaluate it, so a direct experiment would show whether the Pareto-frontier mechanism, not the specific tree learner, is what delivers the trade-off.
  • A natural stress test is to apply GroupCART to datasets with more than two protected-attribute values or to non-binary protected attributes, since the current experiments only cover binary protected attributes.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes GroupCART, an ensemble of fairness-aware decision trees in which each tree is built with a different weight ratio between information gain on the protected attribute (IGS) and information gain on the class label (IGC). Candidate trees are filtered by Pareto-based non-dominated sorting on validation data, and the surviving models vote on test instances. The authors evaluate GroupCART on seven fairness datasets against CART, EGR, Fax-AI, MAAT, and FairMask, reporting accuracy, F1, group fairness metrics (AOD, EOD, SPD, DI), and an individual fairness flip rate (FR). They claim that GroupCART achieves on-par or better predictive performance while substantially improving fairness, and that the method can provide an 'optimal performance-fairness trade-off' based on distance-to-heaven (d2h) scores.

Significance. The central idea, treating fairness as a configurable property of tree-based learners and using multi-objective optimization plus ensemble voting to navigate the fairness-performance trade-off, is relevant to the empirical software engineering community. The paper has notable strengths: it uses seven datasets, repeats experiments 20 times, compares with four baselines, applies non-parametric significance and effect-size testing, and makes code and data publicly available. If the empirical claims are established after correcting the issues below, this would be a useful contribution to fairness-aware ML software. However, the headline evidence for 'optimal trade-off' currently rests on a d2h computation that reverses the semantics of disparate impact, and the individual-fairness result is weakened by an early-stopping rule that uses the reported outcome as its own stopping criterion. These are load-bearing problems in the current manuscript.

major comments (4)
  1. [§5.2, Eq. (4), Table 6] The d2h computation treats DI = 0 as the optimum, but Table 2 defines Disparate Impact as DI = P(Y=1|PA=0)/P(Y=1|PA=1), for which the fair value is 1.0 (100%). The reported values in Table 6 are only consistent with an optimum of 0 for DI: for example, the Adult:Sex CART row has d2h = 124, which requires a DI contribution of (90-0)^2 = 8100 rather than (90-100)^2 = 100. Under the paper's own definition, a lower DI means the unprivileged group receives a smaller fraction of favorable outcomes, i.e., more disparity, not less. Recomputing Adult:Sex and Adult:Race with |DI-100| reverses both GroupCART-versus-CART comparisons. Because Table 6's 'GroupCART ranked first in 6/8 cases' is the main support for the conclusion that GroupCART provides an optimal performance-fairness trade-off, this metric-direction error is load-bearing. Please correct the optimum vector for DI, re-run all d2h and ranking analyses, and revise the Table 5 caption so that it does not state 'for all fairness metrics, smaller is better.'
  2. [§6.3 and Table 5] The early-stopping rule described in §6.3 grows the ensemble until the flip rate FR reaches zero, and the paper then reports FR = 0 as a favorable outcome for GroupCART in Table 5. The manuscript never states whether the FR used for stopping is computed on the validation set or the test set. If the test set is used, the reported perfect individual fairness is a leakage artifact and the selected ensemble sizes (5, 10, 20) are tuned to the reported outcome. If the validation set is used, the test FR = 0 is still not independent evidence, because the ensemble size was selected to drive FR to zero. Please specify exactly which data split feeds the stopping rule, and, if necessary, re-evaluate with a fixed ensemble size chosen before observing test FR.
  3. [§4.3 and §5.2] The domination definitions in Eq. (1)-(3) state that lower objective values y are preferred, but the objectives listed in §5.2 include accuracy and F1, for which higher values are better. The paper does not state that these metrics are negated or converted to error rates before NDSorting is applied in Algorithm 2. If raw accuracy and F1 are passed to a lower-is-better domination rule, the Pareto filter could rank poor-performance models as dominant, which would undermine the selection of candidate trees. Please specify the exact objective transformation used in the implementation of NDSorting.
  4. [Table 4 and §6.1] The row 'Default: Sex' in Table 4 refers to a dataset that is not listed in Table 1 and is not described in §6.1. In addition, the d2h values in Table 4 (0.46-1.14) are on a completely different scale from those in Table 6 (46-124), and the manuscript does not explain whether the two tables use different objective vectors, different normalizations, or different subsets of metrics. This makes the RQ1 result ('up to 51% improvement from optimization') difficult to interpret and reproduce. Please document the datasets and the exact metric vectors used for each d2h calculation.
minor comments (4)
  1. [Table 2] The DI formula has a typo: it reads 'P(Y=1|PA=0]' with a closing bracket instead of a division slash; it should be 'P(Y=1|PA=0)/P(Y=1|PA=1)'.
  2. [Figure 2 caption] The caption contains a grammatical error: 'Presented in the paper be Cruz et al.' should read 'Presented in the paper by Cruz et al.'.
  3. [Tables 5 and 6] The text says Table 5 reports median values over 20 runs, while Table 6 is described as reporting mean d2h scores; the paper should state clearly which statistic is used in each table and why.
  4. [Algorithm 2 and §6.3] The term 'ensemble size N' is used ambiguously: it sometimes refers to the number of initialized candidate trees and sometimes to the size of the final voting ensemble after Pareto filtering. Please define both quantities explicitly.

Circularity Check

2 steps flagged · score 6.0 of 10

FR=0 is used both as the ensemble-size stopping criterion and as the reported individual-fairness outcome, and the d2h rankings treat disparate impact as lower-is-better, contradicting the paper's own ratio definition.

  1. fitted input called prediction [Section 6.3 'Experimental Rig'; RQ2/RQ3, Section 7 (Tables 4-6, Fig. 4)]
    "Hence, in this paper, we keep doubling the size of the ensembles, stopping after the point where FR= 0. In practice, this means that we will generate ensembles of sizes 5,10, and 20 (as the FR decreases to 0 at the size of 20). ... GroupCART always obtains much better FR (flip rate) scores, indicating that GroupCART can usually ensure perfect individual fairness."

    The stopping rule selects the reported ensemble sizes by growing the ensemble until FR=0, and Table 5 then reports GroupCART with FR=0 on every dataset as evidence of 'perfect individual fairness'. The measured outcome is therefore the target used in model selection, not an independent prediction. Section 6.3 does not state that the stopping FR is computed on a validation split distinct from the test FR reported in Table 5; as written, the same metric is used both to fit the ensemble size and to report the headline fairness result, so the FR=0 row is forced by construction.

  2. other [Table 2; Section 5.2 Eq. 4; Table 5 caption; Table 6/RQ3]
    "Disparate Impact (DI) DI=P (Y=1|PA=0]/P [Y=1|PA=1) Similar to SPD but measuring ratio rather than the probability ... For all performance metrics, greater is better; for all fairness metrics, smaller is better."

    Table 2 defines DI as a ratio whose fair value is 1.0, but Table 5 tells readers that all fairness metrics are 'smaller is better', and the d2h values in Table 6 are computed with DI's optimum at 0: CART on Adult:Sex has d2h=124, which matches a DI contribution of (90-0)^2=8100 rather than (90-100)^2=100. By construction, the d2h evaluation rewards lower DI, so the Table 6 ranking that supports the claim that GroupCART 'can provide an optimal performance-fairness trade-off' is generated under an objective opposite to the paper's own DI definition. This is a definitional/construction error in the main evidence, not merely a stylistic choice.

full rationale

The paper is an empirical study, and most of its machinery is not circular: the weighted IGS/IGC tree construction, Pareto-frontier filtering, and ensemble voting are independent algorithmic choices, and the self-citations (e.g., FairMask) are used as baselines rather than as load-bearing support. No imported uniqueness theorem is invoked. However, the central fairness claims contain two construction-level problems. First, the FR early-stopping rule controls the ensemble size until FR=0 and the same FR=0 is then presented as GroupCART's individual-fairness achievement; that 'prediction' reduces to the selection criterion unless a held-out validation/test separation is explicitly stated, which the paper does not do. Second, the d2h objective used to rank methods treats lower DI as better even though the paper defines DI as a ratio where 1.0 is fair; the reported Table 6 distances are only consistent with a 'heaven' of DI=0. Both issues affect the headline conclusion of an optimal performance-fairness trade-off, making the fairness-axis results partially self-fulfilling, while the predictive-performance comparisons and several other fairness metrics remain potentially informative.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

GroupCART is an empirical method rather than a formal derivation, so the central claim rests on dataset assumptions, metric choices, and the integrity of the validation protocol. The most important assumptions are that protected attributes can be used in the split criterion, that the chosen fairness metrics are valid, and that the FR-based early stopping does not leak test information. The free parameters are the ensemble size, the selected weight ratios, and the choice of objectives for Pareto filtering.

free parameters (3)
  • Ensemble size N = 20 (also 5 and 10 explored)
    N is chosen via early stopping on flip rate (FR=0) in Section 6.3. It determines the pool of IGS/IGC weight ratios and the variance of the voting ensemble, so it is a tuned design parameter rather than an input from theory.
  • IGS/IGC weight ratio for selected models = Selected subset of the grid {i/N} using continuous domination on validation
    The 'optimal weight combination' in RQ1 is fit to validation data using the d2h objective. The selected weights are not predicted a priori and vary by dataset and protected attribute.
  • Objective metric set for Pareto filtering = accuracy, F1, AOD, DI
    The choice of which four metrics define the continuous-domination objective is based on preliminary experiments where those metrics showed greater variance, as stated in the Internal Validity threat. This hand-picked set affects which models are selected.
assumptions (6)
  • domain assumption Protected attributes are available in the training data and may be used as split criteria.
    The FDT splitting rule uses information gain on protected attributes (IGS) as an objective, which presumes protected attributes are present and that using them in splitting is an acceptable fairness strategy.
  • domain assumption The fairness metrics AOD, EOD, SPD, DI, and FR as defined in Table 2 are valid measures of bias.
    The paper adopts these metrics from prior literature without defending their normative validity. All conclusions that GroupCART is 'fairer' depend entirely on these definitions.
  • ad hoc to paper The validation set provides an unbiased guide for selecting the Pareto front and early stopping, with no leakage from the test set.
    Section 6.3 uses FR to decide when to stop growing the ensemble but does not state whether that FR is computed on validation or test data. The validity of the test-set fairness claims depends on this unstated procedural assumption.
  • domain assumption The test data are unbiased and correctly labeled.
    Stated in the Conclusion Validity threat (Section 8) and relied on by all evaluation results.
  • ad hoc to paper The d2h Euclidean distance with equal weights is a meaningful aggregate of performance and fairness.
    Equation 4 sums squared deviations across heterogeneous metrics without a stated normalization or utility model. The paper uses d2h both for selecting the optimal weight combination and for evaluating the resulting improvement.
  • ad hoc to paper Continuous domination as defined in Equations 2 and 3 ranks configurations consistently with user preferences.
    The exponential loss in Equation 3 is arbitrary and not derived from a utility model. The Pareto set and the final voting ensemble depend on this particular definition.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Software Engineering Principles for Fairer Systems: Experiments with GroupCART." pith.science (2026). https://pith.science/paper/DHNTZNI4

@misc{pith2026250412587,
  author       = {Pith},
  title        = {Pith review of: Software Engineering Principles for Fairer Systems: Experiments with GroupCART},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DHNTZNI4}},
  note         = {Machine review of arXiv:2504.12587}
}
read the original abstract

Discrimination-aware classification aims to make accurate predictions while satisfying fairness constraints. Traditional decision tree learners typically optimize for information gain in the target attribute alone, which can result in models that unfairly discriminate against protected social groups (e.g., gender, ethnicity). Motivated by these shortcomings, we propose GroupCART, a tree-based ensemble optimizer that avoids bias during model construction by optimizing not only for decreased entropy in the target attribute but also for increased entropy in protected attributes. Our experiments show that GroupCART achieves fairer models without data transformation and with minimal performance degradation. Furthermore, the method supports customizable weighting, offering a smooth and flexible trade-off between predictive performance and fairness based on user requirements. These results demonstrate that algorithmic bias in decision tree models can be mitigated through multi-task, fairness-aware learning. All code and datasets used in this study are available at: https://github.com/anonymous12138/groupCART.

Figures

Figures reproduced from arXiv: 2504.12587 by the authors.

Figure 1
Figure 1. Pareto frontier in the Adult dataset with protected attribute race. That is to say, nowadays more research is conducted aiming to ensure fairness with the least harm to model perfor￾mance. Among various approaches, En￾semble learning is one of the most ex￾plored techniques. State-of-the-art (SOTA) approaches, such as Fair-SMOTE [21] and MAAT [22], use different forms of ensem￾ble learning to achieve an ideal trade-o… view at source ↗
Figure 2
Figure 2. Presented in the paper be Cruz et al. [26], the Pareto frontier of the fairness [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The general flowchart of the GroupCART Algo. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Results for RQ2. Better results have higher fairness and performance (i.e., found top-right). We present the comprehensive comparison in [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]
Figure 5
Figure 5. Figure 5: Results for RQ4. The two figures on the left-hand side present disparate impact [PITH_FULL_IMAGE:figures/full_fig_p022_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 49 canonical work pages

  1. [1]

    URL http://mlr.cs.umass.edu/ml/datasets/ Adult

    (1994) Uci:adult data set. URL http://mlr.cs.umass.edu/ml/datasets/ Adult

    work page 1994

  2. [2]

    URL https://archive.ics

    (2000) Uci:statlog (german credit data) data set. URL https://archive.ics. uci.edu/ml/datasets/Statlog+(German+Credit+Data)

    work page 2000

  3. [3]

    URL https://archive.ics.uci.edu/ml/ datasets/Heart+Disease

    (2001) Uci:heart disease data set. URL https://archive.ics.uci.edu/ml/ datasets/Heart+Disease

    work page 2001

  4. [4]

    URL https://meps.ahrq.gov/ mepsweb/

    (2015) Medical expenditure panel survey. URL https://meps.ahrq.gov/ mepsweb/

    work page 2015

  5. [5]

    URL https://github.com/ propublica/compas-analysis

    (2015) propublica/compas-analysis. URL https://github.com/ propublica/compas-analysis

    work page 2015

  6. [6]

    URL https://www.kaggle.com/c/ bank-marketing-uci

    (2017) Bank marketing uci. URL https://www.kaggle.com/c/ bank-marketing-uci

    work page 2017

  7. [7]

    URL http://tiny.cc/5v6nzz

    (2018) Facebook says it has a tool to detect bias in its artificial intelligence. URL http://tiny.cc/5v6nzz

    work page 2018

  8. [8]

    URLhttps: //www.microsoft.com/en-us/research/group/fate/

    (2018) Fate: Fairness, accountability, transparency, and ethics in ai. URLhttps: //www.microsoft.com/en-us/research/group/fate/

    work page 2018

Show all 66 references
  1. [9]

    URL https://2019.ase-conferences.org/home/ explain-2019

    (2019) Explain 2019. URL https://2019.ase-conferences.org/home/ explain-2019

    2019

  2. [10]

    URL https://fatconference.org/

    (2024) Acm conference on fairness, accountability, and transparency (acm fat*). URL https://fatconference.org/

    2024

  3. [11]

    In: International conference on machine learning, PMLR, pp 60–69

    Agarwal A, Beygelzimer A, Dudík M, Langford J, Wallach H (2018) A reduc- tions approach to fair classification. In: International conference on machine learning, PMLR, pp 60–69

    2018

  4. [12]

    In: Proceedings of the AAAI confer- ence on artificial intelligence, vol 33, pp 1418–1426

    Aghaei S, Azizi MJ, Vayanos P (2019) Learning optimal and fair decision trees for non-discriminative decision-making. In: Proceedings of the AAAI confer- ence on artificial intelligence, vol 33, pp 1418–1426

    2019

  5. [13]

    Agrawal A, Fu W, Chen D, Shen X, Menzies T (2019) How to "dodge" complex software analytics? arXiv preprint arXiv:190201838

    2019

  6. [14]

    (2018) Ai fairness 360: An extensible toolkit for detecting, understanding, and mitigating unwanted algorithmic bias

    Bellamy RK, Dey K, Hind M, Hoffman SC, Houde S, Kannan K, Lohia P, Mar- tino J, Mehta S, Mojsilovic A, et al. (2018) Ai fairness 360: An extensible toolkit for detecting, understanding, and mitigating unwanted algorithmic bias. arXiv preprint arXiv:181001943

    2018

  7. [15]

    In: 2019 IEEE 31st International conference on tools with artificial intelligence (ICTAI), IEEE, pp 1810–1814

    Bhaskaruni D, Hu H, Lan C (2019) Improving prediction fairness via model ensemble. In: 2019 IEEE 31st International conference on tools with artificial intelligence (ICTAI), IEEE, pp 1810–1814

    2019

  8. [16]

    Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering DOI 10.1145/3368089

    Biswas S, Rajan H (2020) Do the machine learning models on a crowd sourced platform exhibit bias? an empirical study on model fairness. Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineerin...

    2020

  9. [17]

    Calmon F, Wei D, Vinzamuri B, Natesan Ramamurthy K, Varshney KR (2017) Optimized pre-processing for discrimination prevention. In: Guyon I, Luxburg UV , Bengio S, Wallach H, Fergus R, Vishwanathan S, Garnett R (eds) Advances in Neural Information Processing Systems 30, Curran ...

    2017

  10. [18]

    1905.05786

    Chakraborty J, Xia T, Fahid FM, Menzies T (2019) Software engineering for fairness: A case study with hyperparameter optimization. 1905.05786

    2019 arXiv

  11. [19]

    Chakraborty J, Majumder S, Yu Z, Menzies T (2020) Fairway: A way to build fair ml software. In: Proceedings of the 28th ACM Joint Meeting on Euro- pean Software Engineering Conference and Symposium on the Foundations of Software Engineering, Association for Computing Machinery...

    2020

  12. [20]

    Chakraborty J, Peng K, Menzies T (2020) Making fair ml software using trust- worthy explanation. In: Proceedings of the 35th IEEE/ACM International Con- ference on Automated Software Engineering, Association for Computing Ma- chinery, New York, NY , USA, ASE ’20, p 1229–1233, ...

    2020

  13. [21]

    Chakraborty J, Majumder S, Menzies T (2021) Bias in machine learning soft- ware: Why? how? what to do? In: Proceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foun- dations of Software Engineering, Association for Computin...

    2021

  14. [22]

    In: Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, pp 1122–1134

    Chen Z, Zhang JM, Sarro F, Harman M (2022) Maat: a novel ensemble approach to addressing fairness and performance bugs for machine learning software. In: Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Eng...

    2022

  15. [23]

    ACM Trans Softw Eng Methodol 32(4), DOI 10.1145/3583561, URL https://doi.org/ 10.1145/3583561

    Chen Z, Zhang JM, Sarro F, Harman M (2023) A comprehensive empirical study of bias mitigation methods for machine learning classifiers. ACM Trans Softw Eng Methodol 32(4), DOI 10.1145/3583561, URL https://doi.org/ 10.1145/3583561

    2023 doi

  16. [24]

    ACM Trans Softw Eng Methodol 33(5), DOI 10.1145/3652155, URL https://doi.org/10.1145/3652155

    Chen Z, Zhang JM, Hort M, Harman M, Sarro F (2024) Fairness testing: A com- prehensive survey and analysis of trends. ACM Trans Softw Eng Methodol 33(5), DOI 10.1145/3652155, URL https://doi.org/10.1145/3652155

    2024 doi

  17. [25]

    Publications Office, DOI doi/10.2759/177365

    Commission E, Directorate-General for Communications Networks C, Tech- nology (2019) Ethics guidelines for trustworthy AI. Publications Office, DOI doi/10.2759/177365

    2019 doi

  18. [26]

    In: 2021 IEEE International Conference on Data Mining (ICDM), IEEE, pp 1036–1041

    Cruz AF, Saleiro P, Belém C, Soares C, Bizarro P (2021) Promoting fairness through hyperparameter optimization. In: 2021 IEEE International Conference on Data Mining (ICDM), IEEE, pp 1036–1041

    2021

  19. [27]

    IEEE transactions on evolutionary computation 6(2):182–197

    Deb K, Pratap A, Agarwal S, Meyarivan T (2002) A fast and elitist multiobjec- tive genetic algorithm: Nsga-ii. IEEE transactions on evolutionary computation 6(2):182–197

    2002

  20. [28]

    In: proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data min- ing, pp 259–268 26 Kewen Peng et al

    Feldman M, Friedler SA, Moeller J, Scheidegger C, Venkatasubramanian S (2015) Certifying and removing disparate impact. In: proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data min- ing, pp 259–268 26 Kewen Peng et al

    2015

  21. [29]

    it’s actually not that clear

    Feller A, Pierson E, Corbett-Davies S, Goel S (2016) A computer program used for bail and sentencing decisions was labeled biased against blacks. it’s actually not that clear. The Washington Post 17

    2016

  22. [30]

    Proceedings of the 2017 11th Joint Meeting on Foundations of Software Engineering - ESEC/FSE 2017 DOI 10.1145/3106237.3106277, URL http://dx.doi.org/10.1145/3106237.3106277

    Galhotra S, Brun Y , Meliou A (2017) Fairness testing: testing software for dis- crimination. Proceedings of the 2017 11th Joint Meeting on Foundations of Software Engineering - ESEC/FSE 2017 DOI 10.1145/3106237.3106277, URL http://dx.doi.org/10.1145/3106237.3106277

    2017

  23. [31]

    In: 2023 IEEE/ACM 45th Inter- national Conference on Software Engineering (ICSE), pp 1533–1545, DOI 10.1109/ICSE48619.2023.00133

    Gohar U, Biswas S, Rajan H (2023) Towards understanding fairness and its composition in ensemble machine learning. In: 2023 IEEE/ACM 45th Inter- national Conference on Software Engineering (ICSE), pp 1533–1545, DOI 10.1109/ICSE48619.2023.00133

    2023

  24. [32]

    In: Proceedings of the 2022 ACM Conference on Fair- ness, Accountability, and Transparency, pp 1905–1916

    Grabowicz PA, Perello N, Mishra A (2022) Marrying fairness and explainability in supervised learning. In: Proceedings of the 2022 ACM Conference on Fair- ness, Accountability, and Transparency, pp 1905–1916

    2022

  25. [33]

    arXiv preprint arXiv:170610208

    Grgi ´c-Hlaˇca N, Zafar MB, Gummadi KP, Weller A (2017) On fairness, diversity and randomness in algorithmic decision making. arXiv preprint arXiv:170610208

    2017

  26. [34]

    1610.02413

    Hardt M, Price E, Srebro N (2016) Equality of opportunity in supervised learn- ing. 1610.02413

    2016 arXiv

  27. [35]

    In: annual meeting of the American Educational Research Association, pp 1–30

    Hess MR, Kromrey JD (2004) Robust confidence intervals for effect sizes: A comparative study of cohen’sd and cliff’s delta under non-normality and hetero- geneous variances. In: annual meeting of the American Educational Research Association, pp 1–30

    2004

  28. [36]

    Scientific american 267(1):66–73

    Holland JH (1992) Genetic algorithms. Scientific american 267(1):66–73

    1992

  29. [37]

    Empirical Software En- gineering 29(1):36

    Hort M, Zhang JM, Sarro F, Harman M (2024) Search-based automatic repair for fairness and accuracy in decision-making software. Empirical Software En- gineering 29(1):36

    2024

  30. [38]

    Knowledge and Information Systems 33(1):1–33

    Kamiran F, Calders T (2012) Data preprocessing techniques for classification without discrimination. Knowledge and Information Systems 33(1):1–33

    2012

  31. [39]

    In: 2010 IEEE international conference on data mining, IEEE, pp 869– 874

    Kamiran F, Calders T, Pechenizkiy M (2010) Discrimination aware decision tree learning. In: 2010 IEEE international conference on data mining, IEEE, pp 869– 874

    2010

  32. [40]

    Inf Sci DOI 10.1016/j.ins.2017

    Kamiran F, Mansha S, Karim A, Zhang X (2018) Exploiting reject option in classification for social discrimination control. Inf Sci DOI 10.1016/j.ins.2017. 09.064

    2018 doi

  33. [41]

    In: Flach PA, De Bie T, Cristianini N (eds) Ma- chine Learning and Knowledge Discovery in Databases, Springer Berlin Heidel- berg, Berlin, Heidelberg, pp 35–50

    Kamishima T, Akaho S, Asoh H, Sakuma J (2012) Fairness-aware classifier with prejudice remover regularizer. In: Flach PA, De Bie T, Cristianini N (eds) Ma- chine Learning and Knowledge Discovery in Databases, Springer Berlin Heidel- berg, Berlin, Heidelberg, pp 35–50

    2012

  34. [42]

    Evolutionary computation

    Laumanns M, Thiele L, Deb K, Zitzler E (2002) Combining convergence and diversity in evolutionary multiobjective optimization. Evolutionary computation

    2002

  35. [43]

    Li Y , Meng L, Chen L, Yu L, Wu D, Zhou Y , Xu B (2022) Training data de- bugging for the fairness of machine learning software. In: Proceedings of the 44th International Conference on Software Engineering, Association for Com- puting Machinery, New York, NY , USA, ICSE ’22, p...

    2022

  36. [44]

    Universitas Psychologica 10(2):545–555

    Macbeth G, Razumiejczyk E, Ledesma RD (2011) Cliff’s delta calculator: A non-parametric effect size program for two groups of observations. Universitas Psychologica 10(2):545–555

    2011

  37. [45]

    IEEE Transactions on software engineering 39(6):822–834

    Menzies T, Butcher A, Marcus A, Cok D, Shull F, Turhan B, Zimmermann T (2013) Local versus global lessons for defect prediction and effort estimation. IEEE Transactions on software engineering 39(6):822–834

    2013

  38. [46]

    Morgan Kaufmann

    Menzies T, Kocaguneli E, Turhan B, Minku L, Peters F (2014) Sharing data and models in software engineering. Morgan Kaufmann

    2014

  39. [47]

    IEEE Transactions on soft- ware engineering 39(4):537–551

    Mittas N, Angelis L (2012) Ranking and clustering software cost estimation models through a multiple comparisons algorithm. IEEE Transactions on soft- ware engineering 39(4):537–551

    2012

  40. [48]

    IEEE Transactions on Software Engineering 46(7):794–811

    Nair V , Yu Z, Menzies T, Siegmund N, Apel S (2018) Finding faster configura- tions using flash. IEEE Transactions on Software Engineering 46(7):794–811

    2018

  41. [49]

    Narayanan A (2019) Tl;ds - 21 fairness definition and their politics by arvind narayanan

    2019

  42. [50]

    IEEE Transactions on Software Engi- neering

    Peng K, Chakraborty J, Menzies T (2022) Fairmask: Better fairness via model- based rebalancing of protected attributes. IEEE Transactions on Software Engi- neering

    2022

  43. [51]

    Empirical Software Engineer- ing 28(3):61

    Peng K, Kaltenecker C, Siegmund N, Apel S, Menzies T (2023) Veer: enhancing the interpretability of model-based optimizations. Empirical Software Engineer- ing 28(3):61

    2023

  44. [52]

    arXiv preprint arXiv:170902012

    Pleiss G, Raghavan M, Wu F, Kleinberg J, Weinberger KQ (2017) On fairness and calibration. arXiv preprint arXiv:170902012

    2017

  45. [53]

    In: Proceedings of the 30th ACM International Con- ference on Information & Knowledge Management, pp 1508–1517

    Ranzato F, Urban C, Zanella M (2021) Fairness-aware training of decision trees by abstract interpretation. In: Proceedings of the 30th ACM International Con- ference on Information & Knowledge Management, pp 1508–1517

    2021

  46. [54]

    In: 2017 IEEE Canada International Humanitarian Tech- nology Conference (IHTC), IEEE, pp 197–201

    Shahriari K, Shahriari M (2017) Ieee standard review—ethically aligned de- sign: A vision for prioritizing human wellbeing with artificial intelligence and autonomous systems. In: 2017 IEEE Canada International Humanitarian Tech- nology Conference (IHTC), IEEE, pp 197–201

    2017

  47. [55]

    Ars Technica 29

    Simonite T (2020) Google offers to help others with the tricky ethics of ai. Ars Technica 29

    2020

  48. [56]

    In: Proceedings of the 44th International Confer- ence on Software Engineering, pp 909–920

    Tizpaz-Niari S, Kumar A, Tan G, Trivedi A (2022) Fairness-aware configuration of machine learning libraries. In: Proceedings of the 44th International Confer- ence on Software Engineering, pp 909–920

    2022

  49. [57]

    (2019) Unlocking fairness: a trade-off revisited

    Wick M, Tristan JB, et al. (2019) Unlocking fairness: a trade-off revisited. Ad- vances in neural information processing systems 32

    2019

  50. [58]

    arXiv preprint arXiv:180500336

    Xia T, Krishna R, Chen J, Mathew G, Shen X, Menzies T (2018) Hyperparameter optimization for effort estimation. arXiv preprint arXiv:180500336

    2018

  51. [59]

    Proc ACM Softw Eng 1(FSE), DOI 10.1145/3660801, URL https://doi.org/10

    Xiao Y , Zhang JM, Liu Y , Mousavi MR, Liu S, Xue D (2024) Mirrorfair: Fixing fairness bugs in machine learning software via counterfactual predictions. Proc ACM Softw Eng 1(FSE), DOI 10.1145/3660801, URL https://doi.org/10. 1145/3660801

    2024 doi

  52. [60]

    Expert Systems with Applications 36(2):2473–2480 28 Kewen Peng et al

    Yeh IC, Lien Ch (2009) The comparisons of data mining techniques for the pre- dictive accuracy of probability of default of credit card clients. Expert Systems with Applications 36(2):2473–2480 28 Kewen Peng et al

    2009

  53. [61]

    1801.07593

    Zhang BH, Lemoine B, Mitchell M (2018) Mitigating unwanted biases with ad- versarial learning. 1801.07593

    2018 arXiv

  54. [62]

    ignorance and prejudice

    Zhang JM, Harman M (2021) "ignorance and prejudice" in software fairness. In: 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE), pp 1436–1447, DOI 10.1109/ICSE43902.2021.00129

    2021

  55. [63]

    In: Proceedings of the ACM/IEEE 42nd International Conference on Software Engineering, pp 949– 960

    Zhang P, Wang J, Sun J, Dong G, Wang X, Wang X, Dong JS, Dai T (2020) White-box fairness testing through adversarial sampling. In: Proceedings of the ACM/IEEE 42nd International Conference on Software Engineering, pp 949– 960

    2020

  56. [64]

    arXiv preprint arXiv:190707237

    Zhang W, Ntoutsi E (2019) Faht: an adaptive fairness-aware decision tree classi- fier. arXiv preprint arXiv:190707237

    2019

  57. [65]

    IEEE transactions on Evolutionary Computation 3(4):257–271

    Zitzler E, Thiele L (1999) Multiobjective evolutionary algorithms: a comparative case study and the strength pareto approach. IEEE transactions on Evolutionary Computation 3(4):257–271

    1999

  58. [66]

    TIK-report 103

    Zitzler E, Laumanns M, Thiele L (2001) Spea2: Improving the strength pareto evolutionary algorithm. TIK-report 103

    2001

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.