Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Gender Fairness of Machine Learning Algorithms for Pain Detection

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Machine-learning pain detectors trained on facial expressions all show gender-based disparities, and even the most accurate one satisfies only some fairness metrics.

desk verdict Useful first fairness benchmark for pain detection on UNBC, but the abstract overclaims algorithmic bias beyond what the subject-leaking, imbalanced protocol can support. read the letter →

arxiv 2506.11132 v1 pith:PQZCYWB3 submitted 2025-06-10 cs.CV cs.LG

Dylan Green , Yuting Shang , Jiaee Cheong , Yang Liu , Hatice Gunes This is my paper
classification cs.CVcs.LG
keywords genderfairnesspaindetectionalgorithmicbiasVisionTransformersupportvectormachinemetricsUNBC-McMasterdatasetfacialexpressionanalysis
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

Automated pain detection from facial expressions could help patients who cannot report pain, but it must treat men and women equally. This paper trains four classifiers — Linear SVM, RBF SVM, CNN, and Vision Transformer — on the UNBC-McMaster shoulder-pain dataset and audits each with seven fairness metrics alongside accuracy. It finds that every model shows gender-based disparities on at least one fairness metric and that no model satisfies all fairness criteria at once. The Vision Transformer reaches the highest accuracy (0.9806) and the best Equal Opportunity and Equalised Odds values, yet it still shows imbalance in Disparate Impact, Demographic Parity, and Treatment Equality. The paper concludes that fairness-aware methods are needed and notes that data imbalance limits conclusive algorithmic-bias claims.

What carries the argument

The key machinery is a seven-metric fairness audit applied to gender-split confusion matrices: Equal Accuracy, Equal Opportunity, Equalised Odds, Disparate Impact, Demographic Parity, Treatment Equality, and Test Fairness, with pain labels derived from the Prkachin–Solomon Pain Intensity formula $\text{PSPI} = AU4 + \max(AU6,AU7) + \max(AU9,AU10) + AU43$. The audit is what turns raw classification errors into a fairness verdict: for each model, accuracy metrics are computed overall and fairness metrics are computed by comparing male and female confusion-matrix entries, so the conclusion that 'no model is fair' rests on this multi-metric comparison.

What would settle it

Rebuild the test set so that male and female groups have the same number of pain and no-pain frames (or reweight the confusion matrices in Table V to equal base rates) and recompute Equalised Odds and Disparate Impact; if the gender gaps shrink to near zero, the reported biases are dataset artifacts rather than algorithmic bias.

Watch

Extended reading notes

Core claim

The paper's central claim is that gender fairness is not achieved by any of the four tested pain-detection classifiers on the UNBC dataset: although ViT achieves the best overall accuracy (0.9806), F1 (0.9808), and ROC AUC (0.9969), every model exhibits measurable disparities between male and female groups under at least one fairness metric, and none satisfies all seven simultaneously. The authors interpret the results as evidence of a persistent accuracy–fairness trade-off in automated pain detection. They also report that ViT is the only model that never lands in the worst position on any fairness metric, which they attribute to its pre-trained, globally attentive architecture; nevertheless, the paper itself cautions that because female test data contained roughly twice as many pain frames as male test data (1132 versus 553), the Disparate Impact and Demographic Parity values do not conclusively establish algorithmic bias.

Load-bearing premise

The load-bearing premise is that the measured fairness-metric gaps are caused by the models themselves rather than by the different numbers of male and female pain images in the test data; the paper itself notes its metrics do not conclusively establish algorithmic bias because female test data contained far more pain frames.

Editorial extensions

If this is right

  • Deploying any of these four classifiers in a clinical pain-assessment setting would treat male and female patients unequally on at least one fairness criterion.
  • ViT is the most promising architecture among those tested: it has the highest accuracy and the best Equal Opportunity and Equalised Odds values, and it is the only model never worst on any fairness metric.
  • Higher accuracy does not automatically bring higher fairness, since the Linear SVM had the best Demographic Parity and Disparate Impact but the lowest accuracy.
  • No single model satisfies all seven fairness metrics simultaneously, matching the known result that fairness metrics can be mutually incompatible.
  • The accuracy–fairness trade-off persists in pain detection, so fairness-aware training or post-hoc mitigation will be needed before these systems are clinically usable.

Reading between the lines

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

  • Because the paper's own test-set numbers show female pain frames outnumber male pain frames by about two to one, a direct extension is that the measured disparities may largely reflect label imbalance rather than model behavior; reweighting the confusion matrices in Table V to equal base rates would test this.
  • The SMOTE oversampling step operated per participant on raw pixels and changed the training gender balance from roughly equal (14,413 male vs. 14,606 female) to skewed (20,630 male vs. 27,384 female), so the 'algorithmic bias' could partly be an artifact of preprocessing rather than the model architectures.
  • A practical deployment rule that follows from this study is to select the fairness metric that matches the clinical harm to avoid (for instance, Equalised Odds to avoid missed pain in either gender) and to report subgroup base rates alongside any fairness metric.
  • A testable extension would be to repeat the same pipeline while stratifying SMOTE by gender so that class balancing does not alter the gender ratio, then compare whether the fairness gaps shrink or disappear.
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

4 major / 5 minor

Summary. This paper compares four classifiers (Linear SVM, RBF SVM, CNN/ResNet-50, and ViT) for binary pain detection from facial images on the UNBC-McMaster Shoulder Pain Expression Archive, reporting accuracy, F1 score, ROC AUC, and seven fairness metrics split by gender. The authors report that ViT achieves the highest accuracy and the best values on several fairness metrics, yet all models show disparities on at least one fairness metric, leading to the abstract claim that "all models exhibited gender-based biases." The paper also acknowledges multiple limitations, including the small dataset, class imbalance, and sensitivity of some fairness metrics to group sizes.

Significance. The topic is important and under-studied: gender fairness of automated pain detection has received little prior attention, and a systematic comparison of traditional ML and deep learning models across multiple fairness metrics would be a useful contribution to affective computing. The paper is transparent in reporting full confusion matrices and in listing fairness metric formulas, which helps reproducibility. However, the central claim of algorithmic gender bias is not currently supported by the experimental design because of subject-level data leakage and demographic confounds introduced during preprocessing. If the authors can reanalyze with a subject-exclusive split and gender-balanced resampling, or carefully restrict the conclusion to the specific skewed dataset, the work could become a meaningful contribution.

major comments (4)
  1. [Section III-A] The data split is performed at the frame level across each participant's images rather than at the subject level; as described, "the split was done 60/20/20 ... across the image dataset" while ensuring that each participant's pain and no-pain images appear in training, validation, and testing. Consequently, frames from the same 25 subjects (13 female, 12 male) occur in both training and test sets, so the gender comparison is effectively a comparison of 13 against 12 individuals, and the fairness gaps reported in Table IV (e.g., ViT Equal Opportunity difference of 0.0268) could arise from identity memorization or subject-specific pain expression rather than from gender-based algorithmic bias. A subject-exclusive split, with a corresponding subject-level resampling strategy, is necessary to support the claim of algorithmic gender bias.
  2. [Section IV-A] The paper itself states that "given the higher proportion of pain instances in the female test data compared to male pain images, these metrics (8) (9) do not provide conclusive evidence of algorithmic bias." Since the test set contains 1132 female pain images versus 553 male pain images, and since Section V reports that SMOTE oversampling increased the gender imbalance to 20,630 male versus 27,384 female training images, the observed disparities in all fairness metrics are confounded by group-specific class priors and training distribution. The abstract's categorical claim that "all models exhibited gender-based biases" therefore overstates what the reported experiments can establish.
  3. [Section V] The paper identifies a limitation that undermines its central claim: "SMOTE oversampling increased gender imbalance (20,630 male vs. 27,384 female images) despite initial balanced distribution (14,413 male vs. 14,606 female)." Because the oversampling was applied to the minority pain class without preserving gender balance, the training data itself contains a demographic confound. The fairness metrics measured at test time therefore cannot separate algorithmic bias from dataset bias, which is the very distinction the paper's introduction (Section I) sets out to investigate. Either the experimental design must be changed to include subject-level and gender-balanced resampling, or the conclusion must be restricted to "models trained on this particular skewed dataset exhibit disparities" rather than "algorithmic bias."
  4. [Table IV] The reported fairness metrics are point estimates without confidence intervals or significance tests. With only 25 subjects, the observed differences (e.g., Equal Accuracy of 0.0027 for RBF SVM versus 0.0098 for CNN) may be within sampling variability. The paper should provide at least bootstrap or permutation-based confidence intervals for the fairness metrics, or otherwise justify that the disparities are statistically meaningful, before concluding that all models exhibit gender-based biases.
minor comments (5)
  1. [Section IV-A] The sentence "However it is important to note debates regarding the reliability of ROC AUC as a performance metric in binary classification" appears twice consecutively; the duplicate should be removed.
  2. [Table III] The ROC AUC formula appears to have a formatting error, rendering as "Z 1 0 TPRdFPR"; it should be the integral of TPR with respect to FPR over [0,1].
  3. [Section III-B] There is a typo in "SMOTEfromimblearn.over sampling": it should read "SMOTE from imblearn.over_sampling".
  4. [Section III-B] The phrase "An example output (2) is shown in Figure 1" references equation (2), which is the accuracy formula; the intended reference is likely to the PSPI equation or to the SMOTE-generated image, and should be corrected.
  5. [Section III-A] There is a spacing/typo issue in "gender.and we used stratified sampling"; it should be "gender. We used stratified sampling".

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the fairness evaluation is an empirical measurement with no fitted prediction or self-referential derivation.

full rationale

This paper is an empirical benchmarking study: four classifiers are trained on the UNBC dataset and evaluated with standard performance and fairness metrics computed from held-out test predictions. None of the reported quantities (Accuracy, F1, ROC AUC, Equal Opportunity, Equalised Odds, Disparate Impact, etc.) is fitted to a target or derived from a parameter that was tuned to reproduce those same metrics. The models are trained and tuned using validation accuracy/loss, and the fairness metrics are then computed from test-set confusion matrices; no equation reduces a fairness metric to the training objective, and no fitted parameter is renamed as a prediction. The self-citations appearing in the literature review (e.g., Cheong, Kalkan, Gunes) serve as background references on fairness in affective computing, not as load-bearing evidence for the empirical results, which are self-contained against the UNBC dataset. The paper itself acknowledges a limitation concerning whether disparities reflect dataset imbalance rather than algorithmic bias, but that is a validity caveat, not circular reasoning. The central claim that all models exhibited gender-based disparities is an interpretation of measured test-set statistics, not a construction that assumes its conclusion. Therefore, no circular step is present and the circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The empirical claims rest on the correctness of the UNBC labels, the choice of fairness metrics, the single data split, and the validity of SMOTE-generated training images. The SMOTE step is particularly consequential because it changed the gender balance of the training set, which directly undermines the attribution of observed disparities to algorithmic bias.

free parameters (5)
  • PSPI binarization threshold = PSPI > 0
    Frames with PSPI score greater than 0 are labeled as pain; this choice defines the class distribution and the fairness evaluation.
  • Data split ratio = 60/20/20
    Images split into train/validation/test per cited best practice; single split, no cross-validation.
  • SMOTE oversampling factor = 4.75 (pain class 5012 to 23798, then duplicated to 24007)
    Chosen to balance classes; increased gender imbalance in training data from 14,413:14,606 to 20,630:27,384.
  • Random seed = 42
    Controls all preprocessing and training; single seed, no repeated runs or variance estimates.
  • Model hyperparameters = See Section III-C/D: e.g., alpha=0.001, gamma=0.001, lr=0.0001, wd=0.1
    Tuned via grid search on validation; final values affect all reported metrics.
assumptions (4)
  • domain assumption UNBC dataset PSPI scores and gender labels are ground truth
    All performance and fairness evaluations treat PSPI scores and manual gender annotations as correct.
  • domain assumption The seven fairness metrics operationalize gender fairness appropriately
    Table III metrics are used without justification that they are valid for pain detection in a clinical fairness context.
  • domain assumption The single 60/20/20 split is representative
    All conclusions are based on one split of 25 subjects; no repeated k-fold or bootstrapping is performed.
  • domain assumption SMOTE-generated and augmented images are valid for fairness evaluation
    Synthetic pain images are used in training without verification that they preserve gender-specific facial appearance or do not introduce artifacts affecting fairness.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Gender Fairness of Machine Learning Algorithms for Pain Detection." pith.science (2026). https://pith.science/paper/PQZCYWB3

@misc{pith2026250611132,
  author       = {Pith},
  title        = {Pith review of: Gender Fairness of Machine Learning Algorithms for Pain Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PQZCYWB3}},
  note         = {Machine review of arXiv:2506.11132}
}
read the original abstract

Automated pain detection through machine learning (ML) and deep learning (DL) algorithms holds significant potential in healthcare, particularly for patients unable to self-report pain levels. However, the accuracy and fairness of these algorithms across different demographic groups (e.g., gender) remain under-researched. This paper investigates the gender fairness of ML and DL models trained on the UNBC-McMaster Shoulder Pain Expression Archive Database, evaluating the performance of various models in detecting pain based solely on the visual modality of participants' facial expressions. We compare traditional ML algorithms, Linear Support Vector Machine (L SVM) and Radial Basis Function SVM (RBF SVM), with DL methods, Convolutional Neural Network (CNN) and Vision Transformer (ViT), using a range of performance and fairness metrics. While ViT achieved the highest accuracy and a selection of fairness metrics, all models exhibited gender-based biases. These findings highlight the persistent trade-off between accuracy and fairness, emphasising the need for fairness-aware techniques to mitigate biases in automated healthcare systems.

Figures

Figures reproduced from arXiv: 2506.11132 by the authors.

Figure 1
Figure 1. From left to right, the images represent: (1) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. HOG feature extraction. From left to right, the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. FAIR_XAI: Improving Multimodal Foundation Model Fairness via Explainability for Wellbeing Assessment

    cs.AI 2026-04 unverdicted novelty 5.0 of 10

    Zero-shot vision-language models are unreliable and vary widely for depression screening, and explainability-based fairness interventions often trade away accuracy without reliable fairness gains.

Reference graph

Works this paper leans on

91 extracted references · 67 canonical work pages · cited by 1 Pith paper

  1. [1]

    Aggarwal, P

    A. Aggarwal, P. Lohia, S. Nagar, K. Dey, and D. Saha. Black box fair- ness testing of machine learning models. In Proceedings of the 2019 27th ACM joint meeting on european software engineering conference and symposium on the foundations of software engineering, pages 625–635, 2019

    2019

  2. [2]

    D. E. Attoe and E. A. Climie. Miss. diagnosis: a systematic review of adhd in adult women. Journal of attention disorders, 27(7):645–657, 2023

    2023

  3. [3]

    Barocas, M

    S. Barocas, M. Hardt, and A. Narayanan. Fairness and Machine Learning: Limitations and Opportunities. FairnessML, 2019

    2019

  4. [4]

    Bernhardt, C

    M. Bernhardt, C. Jones, and B. Glocker. Potential sources of dataset bias complicate investigation of underdiagnosis by machine learning algorithms. Nature Medicine, 28(6):1157–1158, 2022

    2022

  5. [5]

    Breivik, P.-C

    H. Breivik, P.-C. Borchgrevink, S.-M. Allen, L.-A. Rosseland, L. Ro- mundstad, E. Breivik Hals, G. Kvarstein, and A. Stubhaug. Assessment of pain. British journal of anaesthesia, 101(1), 2008

    2008

  6. [6]

    Buijsman

    S. Buijsman. Navigating fairness measures and trade-offs. AI and Ethics, 2023

    2023

  7. [7]

    Cameron, J

    J. Cameron, J. Cheong, M. Spitale, and H. Gunes. Multimodal gender fairness in depression prediction: Insights on data from the usa & china. arXiv, 08 2024

    2024

  8. [8]

    Castelnovo, R

    A. Castelnovo, R. Crupi, G. Greco, D. Regoli, I. G. Penco, and A. C. Cosentini. The zoo of fairness metrics in machine learning. 2021

    2021

Show all 91 references
  1. [9]

    Caton and C

    S. Caton and C. Haas. Fairness in machine learning: A survey. ACM Computing Surveys, 55(3):1–35, 2023

    2023

  2. [10]

    S. Y . Chaganti, I. Nanda, K. R. Pandi, T. G. Prudhvith, and N. Kumar. Image classification using svm and cnn. In 2020 International Conference on Computer Science, Engineering and Applications (ICCSEA), pages 1–5, 2020

    2020

  3. [11]

    M. A. Chandra and S. Bedi. Survey on svm and their application in image classification. International Journal of Information Technology, 13(5):1–11, 2021

    2021

  4. [12]

    Chapelle, P

    O. Chapelle, P. Haffner, and V . N. Vapnik. Support vector machines for histogram-based image classification. IEEE transactions on Neural Networks, 10(5):1055–1064, 1999

    1999

  5. [13]

    J. Cheong. Fairness for affective and wellbeing computing. PhD thesis, 2025

    2025

  6. [14]

    Cheong, A

    J. Cheong, A. Bangar, S. Kalkan, and H. Gunes. U-fair: Uncertainty- based multimodal multitask learning for fairer depression detection. In Proceedings of the 4th Machine Learning for Health Symposium, volume 259 of Proceedings of Machine Learning Research, pages 203–

  7. [15]

    Cheong, N

    J. Cheong, N. Churamani, L. Guerdan, T. E. Lee, Z. Han, and H. Gunes. Causal-hri: Causal learning for human-robot interaction. In Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction, pages 1311–1313, 2024

    2024

  8. [16]

    Cheong, S

    J. Cheong, S. Kalkan, and H. Gunes. The hitchhiker’s guide to bias and fairness in facial affective signal processing: Overview and techniques. IEEE Signal Processing Magazine, 38(6):39–49, 2021

    2021

  9. [17]

    Cheong, S

    J. Cheong, S. Kalkan, and H. Gunes. Causal structure learning of bias for fair affect recognition. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 340– 349, 2023

    2023

  10. [18]

    Cheong, S

    J. Cheong, S. Kalkan, and H. Gunes. Counterfactual fairness for facial expression recognition. In ECCV 2022 Workshops, pages 245–261. Springer, 2023

    2022

  11. [19]

    Cheong, S

    J. Cheong, S. Kalkan, and H. Gunes. Fairrefuse: referee-guided fusion for multimodal causal fairness in depression detection. In Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, pages 7224–7232, 2024

    2024

  12. [20]

    Cheong, S

    J. Cheong, S. Kuzucu, S. Kalkan, and H. Gunes. Towards gender fair- ness for mental health prediction. In Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, pages 5932– 5940, 2023

    2023

  13. [21]

    it’s not fair!

    J. Cheong, M. Spitale, and H. Gunes. “it’s not fair!”–fairness for a small dataset of multi-modal dyadic mental well-being coaching. In 2023 11th International Conference on Affective Computing and Intelligent Interaction (ACII), pages 1–8. IEEE, 2023

    2023

  14. [22]

    Cheong, M

    J. Cheong, M. Spitale, and H. Gunes. Small but fair! fairness for multimodal human-human and robot-human mental wellbeing coaching. arXiv preprint arXiv:2407.01562, 2024

    2024 arXiv

  15. [23]

    Chicco and G

    D. Chicco and G. Jurman. The advantages of the matthews correlation coefficient (mcc) over f1 score and accuracy in binary classification evaluation. BMC genomics, 21:1–13, 2020

    2020

  16. [24]

    Chicco and G

    D. Chicco and G. Jurman. The matthews correlation coefficient (mcc) should replace the roc auc as the standard metric for assessing binary classification. BioData Mining, 16(1):4, 2023

    2023

  17. [25]

    Churamani, J

    N. Churamani, J. Cheong, S. Kalkan, and H. Gunes. Towards causal replay for knowledge rehearsal in continual learning. In AAAI Bridge Program on Continual Causality, pages 63–70. PMLR, 2023

    2023

  18. [26]

    Churamani, O

    N. Churamani, O. Kara, and H. Gunes. Domain-incremental continual learning for mitigating bias in facial expression and action unit recognition. IEEE Transactions on Affective Computing, 14(4):3191– 3206, 2022

    2022

  19. [27]

    H. S. Dadi and G. M. Pillutla. Improved face recognition rate using hog features and svm classifier. IOSR Journal of Electronics and Communication Engineering, 11(04):34–44, 2016

    2016

  20. [28]

    Doshi-Velez and B

    F. Doshi-Velez and B. Kim. Towards a rigorous science of interpretable machine learning. arXiv preprint arXiv:1702.08608, 2017

    2017 arXiv

  21. [29]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020

    2010 arXiv

  22. [30]

    Elor and H

    Y . Elor and H. Averbuch-Elor. To smote, or not to smote? arXiv preprint arXiv:2201.08528, 2022

    2022 arXiv

  23. [31]

    Elreedy and A

    D. Elreedy and A. F. Atiya. A comprehensive analysis of synthetic minority oversampling technique (smote) for handling class imbalance. Information Sciences, 505:32–64, 2019

    2019

  24. [32]

    Escrib `a-Ag¨uir and L

    V . Escrib `a-Ag¨uir and L. Artazcoz. Gender differences in postpartum depression: a longitudinal cohort study. Journal of Epidemiology & Community Health, 65(4):320–326, 2011

    2011

  25. [33]

    T. Feng, R. Hebbar, N. Mehlman, X. Shi, A. Kommineni, S. Narayanan, et al. A review of speech-centric trustworthy machine learning: Privacy, safety, and fairness. APSIPA Transactions on Signal and Information Processing, 12(3), 2023

    2023

  26. [34]

    D. Foley. Considerations of sample and feature size. IEEE Transactions on Information Theory, 18(5):618–626, 1972

    1972

  27. [35]

    V . C. Gandhi and P. P. Gandhi. A survey-insights of ml and dl in health domain. In 2022 International Conference on Sustainable Computing and Data Communication Systems (ICSCDS), pages 239–246. IEEE, 2022

    2022

  28. [36]

    P. Garg, J. Villasenor, and V . Foggo. Fairness metrics: A comparative analysis. In 2020 IEEE international conference on big data (Big Data), pages 3662–3666. IEEE, 2020

    2020

  29. [37]

    Gkikas and M

    S. Gkikas and M. Tsiknakis. Automatic assessment of pain based on deep learning methods: A systematic review. Computer methods and programs in biomedicine, 231:107365, 2023

    2023

  30. [38]

    Hardt, E

    M. Hardt, E. Price, and N. Srebro. Equality of opportunity in super- vised learning. Advances in neural information processing systems, 29, 2016

    2016

  31. [39]

    Hassan, D

    T. Hassan, D. Seus, J. Wollenberg, K. Weitz, M. Kunz, S. Lauten- bacher, J. U. Garbas, and U. Schmid. Automatic detection of pain from facial expressions: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(6):1815–1831, 2021

    2021

  32. [40]

    G. A. Hawker, S. Mian, T. Kendzerska, and M. French. Measures of adult pain: Visual analog scale for pain (vas pain), numeric rating scale for pain (nrs pain), mcgill pain questionnaire (mpq), short-form mcgill pain questionnaire (sf-mpq), chronic pain grade scale (cpgs), shor...

    2011

  33. [41]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    2016

  34. [42]

    Hendrycks, S

    D. Hendrycks, S. Basart, N. Mu, S. Kadavath, F. Wang, E. Dorundo, R. Desai, T. Zhu, S. Parajuli, M. Guo, et al. The many faces of robustness: A critical analysis of out-of-distribution generalization. In Proceedings of the IEEE/CVF international conference on computer vision, ...

    2021

  35. [43]

    Hendrycks, X

    D. Hendrycks, X. Liu, E. Wallace, A. Dziedzic, R. Krishnan, and D. Song. Pretrained transformers improve out-of-distribution robust- ness. arXiv preprint arXiv:2004.06100, 2020

    2004 arXiv

  36. [44]

    M. A. Hussain and L. Gogoi. Performance analyses of five neural network classifiers on nodule classification in lung ct images using weka: a comparative study. Physical and Engineering Sciences in Medicine, 45(4):1193–1204, 2022

    2022

  37. [45]

    Hutson and J

    P. Hutson and J. Hutson. Autism in females: Understanding the overlooked diagnoses, unique challenges, and recommendations, 2023

    2023

  38. [46]

    M. P. Jensen, P. Karoly, and S. Braver. Interpretation of visual analog scale ratings and change scores: A reanalysis of two clinical trials of postoperative pain. The Journal of Pain, 4(7):407–414, 2003

    2003

  39. [47]

    G. Jeon. Color image enhancement by histogram equalization in heterogeneous color space. Int. J. Multimedia Ubiquitous Eng, 9(7):309–318, 2014

    2014

  40. [48]

    Kleinberg, S

    J. Kleinberg, S. Mullainathan, and M. Raghavan. Inherent trade-offs in the fair determination of risk scores. arXiv preprint arXiv:1609.05807, 2016

    2016 arXiv

  41. [49]

    Kumar, S

    P. Kumar, S. Happy, and A. Routray. A real-time robust facial ex- pression recognition system using hog features. In 2016 International Conference on Computing, Analytics and Security Trends (CAST), pages 289–293. IEEE, 2016

    2016

  42. [50]

    Kuzucu, J

    S. Kuzucu, J. Cheong, H. Gunes, and S. Kalkan. Uncertainty as a fairness measure. Journal of Artificial Intelligence Research, 81:307– 335, 2024

    2024

  43. [51]

    A. M. H. Kwok, J. Cheong, S. Kalkan, and H. Gunes. Machine learning fairness for depression detection using eeg data. arXiv preprint arXiv:2501.18192, 2025

    2025 arXiv

  44. [52]

    LeCun, Y

    Y . LeCun, Y . Bengio, and G. Hinton. Deep learning. Nature, 521, 2015

    2015

  45. [53]

    M. Li, Y . Wang, C. Yang, Z. Lu, and J. Chen. Automatic diagnosis of depression based on facial expression information and deep convo- lutional neural network. IEEE Transactions on Computational Social Systems, 11(5):5728–5739, 2024

    2024

  46. [54]

    Lopez Martinez, R

    D. Lopez Martinez, R. Picard, et al. Personalized automatic estimation of self-reported pain intensity from facial expressions. In Proceedings of the IEEE conference on computer vision and pattern recognition workshops, pages 70–79, 2017

    2017

  47. [55]

    Loshchilov and F

    I. Loshchilov and F. Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  48. [56]

    Lucey, J

    P. Lucey, J. F. Cohn, K. M. Prkachin, P. E. Solomon, and I. Matthews. Painful data: The unbc-mcmaster shoulder pain expression archive database. In 2011 IEEE International Conference on Automatic Face & Gesture Recognition (FG), pages 57–64, 2011

    2011

  49. [57]

    Mehrabi, F

    N. Mehrabi, F. Morstatter, N. Saxena, K. Lerman, and A. Galstyan. A survey on bias and fairness in machine learning. ACM computing surveys (CSUR), 54(6):1–35, 2021

    2021

  50. [58]

    Micikevicius, S

    P. Micikevicius, S. Narang, J. Alben, G. Diamos, E. Elsen, D. Garcia, B. Ginsburg, M. Houston, O. Kuchaiev, G. Venkatesh, et al. Mixed precision training. arXiv preprint arXiv:1710.03740, 2018

    2018 arXiv

  51. [59]

    I. Muraina. Ideal dataset splitting ratios in machine learning algo- rithms: general concerns for data scientists and data analysts. In 7th international Mardin Artuklu scientific research conference, pages 496–504, 2022

    2022

  52. [60]

    Narkhede

    S. Narkhede. Understanding auc-roc curve. Towards data science, 26(1):220–227, 2018

    2018

  53. [61]

    Nerella, Z

    S. Nerella, Z. Guan, A. Davidson, Y . Ren, T. Baslanti, B. Armfield, P. Tighe, A. Bihorac, and P. Rashidi. Detecting visual cues in the intensive care unit and association with patient clinical status. arXiv preprint arXiv:2311.00565, 2023

    2023 arXiv

  54. [62]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Van- derplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine learning in Python. Journal of Machine Lear...

    2011

  55. [63]

    Pessach and E

    D. Pessach and E. Shmueli. A review on fairness in machine learning. ACM Computing Surveys (CSUR), 55(3):1–44, 2022

    2022

  56. [64]

    K. M. Prkachin and P. E. Solomon. The structure, reliability and validity of pain expression: Evidence from patients with shoulder pain. Pain, 139(2):267–274, 2008

    2008

  57. [65]

    Rodrigo, C

    M. Rodrigo, C. Cuevas, and N. Garc ´ıa. Comprehensive comparison between vision transformers and convolutional neural networks for face recognition tasks. Scientific Reports, 14(1):21392, 2024

    2024

  58. [66]

    S. D. Roy, M. K. Bhowmik, P. Saha, and A. K. Ghosh. An approach for automatic pain detection through facial expression. Procedia Computer Science, 84:99–106, 2016

    2016

  59. [67]

    M. Salama. Optimization of regression models using machine learning: A comprehensive study with scikit-learn. Optimization of Regression Models Using Machine Learning: A Comprehensive Study with Scikit-learn— IUSRJ, 5, 2024

    2024

  60. [68]

    brave men

    A. Samulowitz, I. Gremyr, E. Eriksson, and G. Hensing. “brave men” and “emotional women”: A theory-guided literature review on gender bias in health care and gendered norms towards patients with chronic pain. Pain research and management, 2018(1):6358624, 2018

    2018

  61. [69]

    Saqib, A

    K. Saqib, A. F. Khan, and Z. A. Butt. Machine learning methods for predicting postpartum depression: scoping review. JMIR mental health, 8(11):e29838, 2021

    2021

  62. [70]

    Sch ¨olkopf

    B. Sch ¨olkopf. Learning with kernels: support vector machines, regularization, optimization, and beyond, 2002

    2002

  63. [71]

    Shorten and T

    C. Shorten and T. M. Khoshgoftaar. A survey on image data augmentation for deep learning. Journal of big data, 6(1):1–48, 2019

    2019

  64. [72]

    Shuang, G

    Y . Shuang, G. Liangbo, Z. Huiwen, L. Jing, C. Xiaoying, S. Siyi, Z. Xiaoya, and L. Wen. Classification of pain expression images in elderly with hip fractures based on improved resnet50 network. Frontiers in Medicine, 11, 2024

    2024

  65. [73]

    Spitale, J

    M. Spitale, J. Cheong, and H. Gunes. Underneath the numbers: Quantitative and qualitative gender fairness in llms for depression prediction. arXiv preprint arXiv:2406.08183, 2024

    2024 arXiv

  66. [74]

    W. Sun, O. Nasraoui, and P. Shafto. Evolution and impact of bias in human and machine learning algorithm interaction. Plos one, 15(8):e0235502, 2020

    2020

  67. [75]

    Susan and A

    S. Susan and A. Kumar. The balancing trick: Optimized sampling of imbalanced datasets—a brief survey of the recent state of the art. Engineering Reports, 3(4):e12298, 2021

    2021

  68. [76]

    Tyagi and S

    S. Tyagi and S. Mittal. Sampling approaches for imbalanced data classification problem in machine learning. In Proceedings of ICRIC 2019: Recent innovations in computing, pages 209–221. Springer, 2020

    2019

  69. [77]

    Varona and J

    D. Varona and J. L. Su ´arez. Discrimination, bias, fairness, and trustworthy ai. Applied Sciences, 12(12):5826, 2022

    2022

  70. [78]

    K. R. Vashney. Trustworthy machine learning. Independently pub- lished, 2022

    2022

  71. [79]

    M. B. Vela, A. I. Erondu, N. A. Smith, M. E. Peek, J. N. Woodruff, and M. H. Chin. Eliminating explicit and implicit biases in health care: evidence and research needs. Annual review of public health, 43(1):477–501, 2022

    2022

  72. [80]

    Verma and J

    S. Verma and J. Rubin. Fairness definitions explained. In Proceedings of the international workshop on software fairness, pages 1–7, 2018

    2018

  73. [81]

    R. Wang, P. Chaudhari, and C. Davatzikos. Bias in machine learning models can be significantly mitigated by careful training: Evidence from neuroimaging studies. Proceedings of the National Academy of Sciences, 120(6):e2211613120, 2023

    2023

  74. [82]

    Werner, D

    P. Werner, D. Lopez-Martinez, S. Walter, A. Al-Hamadi, S. Gruss, and R. W. Picard. Automatic recognition methods supporting pain assessment: A survey. IEEE Transactions on Affective Computing, 13(1):530–552, 2019

    2019

  75. [83]

    B. D. Winslow, R. Kwasinski, K. Whirlow, E. Mills, J. Hullfish, and M. Carroll. Automatic detection of pain using machine learning. Frontiers in pain research, 3:1044518, 2022

    2022

  76. [84]

    C. Y . Wong, G. Jiang, M. A. Rahman, S. Liu, S. C.-F. Lin, N. Kwok, H. Shi, Y .-H. Yu, and T. Wu. Histogram equalization and optimal profile compression based approach for colour image enhancement. Journal of Visual Communication and Image Representation, 38:802– 813, 2016

    2016

  77. [85]

    T. Xu, J. White, S. Kalkan, and H. Gunes. Investigating bias and fairness in facial expression recognition. In Computer Vision–ECCV 2020 Workshops: Glasgow, UK, August 23–28, 2020, Proceedings, Part VI 16, pages 506–523. Springer, 2020

    2020

  78. [86]

    J. Yang, A. A. Soltan, D. W. Eyre, Y . Yang, and D. A. Clifton. An adversarial training framework for mitigating algorithmic biases in clinical machine learning. NPJ digital medicine, 6(1):55, 2023

    2023

  79. [87]

    Yang, Y .-F

    T. Yang, Y .-F. Li, M. Mahdavi, R. Jin, and Z.-H. Zhou. Nystr ¨om method vs random fourier features: A theoretical and empirical com- parison. Advances in neural information processing systems, 25, 2012

    2012

  80. [88]

    M. B. Zafar, I. Valera, M. Gomez Rodriguez, and K. P. Gummadi. Fairness beyond disparate treatment & disparate impact: Learning classification without disparate mistreatment. In Proceedings of the 26th international conference on world wide web, pages 1171–1180, 2017

    2017

  81. [89]

    B. H. Zhang, B. Lemoine, and M. Mitchell. Mitigating unwanted bi- ases with adversarial learning. In Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, pages 335–340, 2018

    2018

  82. [90]

    Zhang, X

    X. Zhang, X. Li, Y . Feng, and Z. Liu. The use of roc and auc in the validation of objective image fusion evaluation metrics. Signal processing, 115:38–48, 2015

    2015

  83. [218]

    PMLR, 15–16 Dec 2025

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.