Shapley Values for Feature Selection: The Good, the Bad, and the Axioms
Open accessThe DOI matches its Crossref record; 63 of the 64 references it cites were found in a registry or an index.
Abstract
The Shapley value has become popular in the Explainable AI (XAI) literature, thanks, to a large extent, to a solid theoretical foundation, including four “favourable and fair” axioms for attribution in transferable utility games. The Shapley value is provably the only solution concept satisfying these axioms. In this paper, we introduce the Shapley value and draw attention to its recent uses as a feature selection tool. We call into question this use of the Shapley value, using simple, abstract “toy” counterexamples to illustrate that the axioms may work against the goals of feature selection. From this, we develop a number of insights that are then investigated in concrete simulation settings, with a variety of Shapley value formulations, including SHapley Additive exPlanations (SHAP) and Shapley Additive Global importancE (SAGE). The aim is not to encourage any use of the Shapley value for feature selection, but we aim to clarify various limitations around their current use in the literature. In so doing, we hope to help demystify certain aspects of the Shapley value axioms that are viewed as “favourable”. In particular, we wish to highlight that the favourability of the axioms depends non-trivially on the way in which the Shapley value is appropriated in the XAI application.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“we incorporated the use of interpretability tools (eg, SHAP 28 values, feature importance) to ensure transparency, a priority emphasized in recent ML reporting guidelines.”
From Linguistic markers for identifying post-traumatic stress disorder and associated symptoms: a systematic literature review (2025) · cites this paper as [28]
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“To align with best practices 27 for systematic reviews of ML studies, we augmented PROBAST with criteria tailored to these challenges”
From Linguistic markers for identifying post-traumatic stress disorder and associated symptoms: a systematic literature review (2025) · cites this paper as [28]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
A Unified Approach to Interpreting Model Predictions
Lundberg, S., & Lee, S.-I. (2017) · arXiv · doi:10.48550/arxiv.1705.07874 · Cited by 23,707
Gauriat, C.-M., Pencolé, Y., Ribot, P., & Brouillet, G. (2024) · HAL (Le Centre pour la Communication Scientifique Directe) · doi:10.4230/oasics.dx.2024.27 · Cited by 12,586
From local explanations to global understanding with explainable AI for trees
Lundberg, S. M., Erion, G., Chen, H., DeGrave, A., Prutkin, J. M., Nair, B., Katz, R., Himmelfarb, J., Bansal, N., & Lee, S.-I. (2020) · Nature Machine Intelligence · doi:10.1038/s42256-019-0138-9 · Cited by 10,551
17. A Value for n-Person Games
Shapley, L. S. (1953) · Contributions to the Theory of Games (AM-28), Volume II · doi:10.1515/9781400881970-018 · Cited by 6,950
A survey on feature selection methods
Chandrashekar, G., & Sahin, F. (2014) · Computers & Electrical Engineering · doi:10.1016/j.compeleceng.2013.11.024 · Cited by 5,189
Shapley, L. S. (1988) · The Shapley Value · doi:10.1017/cbo9780511528446.003 · Cited by 4,954
Explaining prediction models and individual predictions with feature contributions
Štrumbelj, E., & Kononenko, I. (2014) · Knowledge and Information Systems · doi:10.1007/s10115-013-0679-x · Cited by 2,286
Tukey, J. W. (1962) · The Annals of Mathematical Statistics · doi:10.1214/aoms/1177704711 · Cited by 1,860
Consistent Individualized Feature Attribution for Tree Ensembles
Lundberg, S. M., Erion, G. G., & Lee, S.-I. (2018) · arXiv · doi:10.48550/arxiv.1802.03888 · Cited by 1,430
Budescu, D. V. (1993) · Psychological Bulletin · doi:10.1037/0033-2909.114.3.542 · Cited by 1,193
Research works printed but not linked: 1
- 18. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, Eds., vol. 30. Curran Associates, Inc., 2017.
Cited by
Items 301–347 of 347
Mukherjee, J., & Dell’Acqua, F. (2024) · IEEE India Geoscience and Remote Sensing Symposium (InGARSS) · doi:10.1109/ingarss61818.2024.10984099
“The impacts of the vegetation indices are studied by Shapley Additive Explanations ( SHAP ), which computes a consistent and locally accurate measure of the contribution of each feature based on cooperative game theory, i.e. Shapley value [8].”
Alothman, H. A. T., Lejmi, W., Ameur, S., & Mahjoub, M. A. (2025) · International Conference on Information Networking · doi:10.1109/icoin63865.2025.10992955
“SHAP works by assigning a “Shapley value” to each feature, which represents its contribution to the final prediction [18].”
Pattern-Based Graph Classification: Comparison of Quality Measures and Importance of Preprocessing
Potin, L., Figueiredo, R., Labatut, V., & Largeron, C. (2025) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/3743143 · Cited by 2
“The SV was later used in machine learning, to perform feature selection, as it allows measuring how much individual features contribute to solving the task at hand [31].”
Wang, M. (2025) · IEEE Access · doi:10.1109/access.2025.3584923
“Meanwhile, feature selection, as a data preprocessing process, has an important position in data mining and machine learning, etc [11].”
Balancing Accuracy and Explainability: An Ensemble-KAN Model for Patent Grant Prediction
Shi, J., Peng, X., Qiao, X., Ming, Y., & Liu, X. (2025) · Scientometrics · doi:10.1007/s11192-026-05639-5
An Effective Hybrid Sampling Strategy for Single-Split Evaluation of Classifiers
Yen, S.-J., Lee, Y., & Tang, Y.-J. (2025) · Electronics · doi:10.3390/electronics14142876
Energy-Based Model for Accurate Estimation of Shapley Values in Feature Attribution
Lü, C., Zeng, J., Xia, Y. C., Cai, J., & Luo, S. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3626404 · Cited by 2
“Although the utilization of Shapley values for explaining DL models may sometimes lead to non-intuitive feature attributions [8], [9], [10], the ideal mathematical properties make it possible to fairly measure the contribution of each input in the model.”
Hulskemper, D. C., Antolínez, J. A. Á., Lindenbergh, R., & Anders, K. (2026) · Earth Surface Dynamics · doi:10.5194/esurf-14-329-2026 · Cited by 1
“Shapley values (Fryer et al., 2021)) and a labelled dataset might increase effectiveness of the clustering.”
Verdeyme, A., Grainger, J. P., Lucia, M. D., & Olhede, S. C. (2025) · medRxiv · doi:10.1101/2025.11.21.25340725
“It is important to note that both SHAP and SAGE values face challenges -such as problems with correlated features (see [46]) - and each method has its inherent limitations.”
An AI-aided carbon conversion framework for efficient carbon storage
Katterbauer, K., Qasim, A., & Ghamdi, A. A. (2025) · Journal of King Saud University - Engineering Sciences · doi:10.1007/s44444-025-00051-5 · Cited by 2
Feature Selection for Load Forecasting based on Shapley Value Estimation
Singh, A., Kiran, D., & Padhy, N. P. (2025) · IEEE International Conference on Sustainable Energy and Future Electric Transportation (SEFET) · doi:10.1109/sefet65155.2025.11255297
“The Shapley value, derived from cooperative game theory, ensures a fair distribution of rewards or contributions among players based on their individual contribution to the overall outcome [19].”
Contrastive Scenarios: Enhancing ML Model Interpretability
Chaudhary, S. (2026) · Lecture notes in networks and systems · doi:10.1007/978-981-96-7140-3_22
Explainable Machine Learning for Transparent Decision-Making in E-Governance Systems
Ghosh, S., Shukla, V. K., Chanda, P., & Singh, A. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-12232-2_19 · Cited by 1
Safii, M., Husain, H., & Marzuki, K. (2025) · Matrik Jurnal Manajemen Teknik Informatika dan Rekayasa Komputer · doi:10.30812/matrik.v25i1.5133 · Cited by 1
“Each feature is multiplied by the corresponding weight, and then the results are summed with the existing bias [16].”
Explainable Machine Learning Based BioDiesel Synthesis from Used Frying Oil
Nguyen, T. T. H., Cao, D. N., Truong, T. H., Nguyen, A. Q., & Trần, V. D. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-14194-1_1
de Souza, R. M. M., Souza, L. V. B., Leal, D. A., Biczyk, M., Amorim, M. M., & de Castro, L. N. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-04160-9_33
Near-instantaneous Cardiovascular Event Prediction Using Multimodal Deep Learning
Al-Hasooni, M., Abdzaid, A., Hadi, N., & Fakhruldeen, H. (2026) · Electronics ETF · doi:10.53314/els2630003a
Prediction of the Development Height of Fractured Water-Conducting Zones in Chinese Coal Mines
Ji, Y., Yu, L., Dong, D., & Wei, Z. (2026) · Mine Water and the Environment · doi:10.1007/s10230-026-01101-3
“The strength of SHAP values lies in their strong interpretability, offering transparent and intuitive explanations of the model's predictions, which facilitates model analysis and optimization (Fryer et al. 2021; Ning et al. 2022).”
Multi-choice Solutions for Feature and Parameter Importance
Fryer, D., Lowing, D., Strümke, I., & Nguyen, H. D. (2026) · Communications in computer and information science · doi:10.1007/978-981-95-6888-8_4 · Cited by 1
Self-citationBeyond additivity: sparse isotonic shapley regression toward nonlinear explainability
She, J. (2026) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-026-01196-w
“In particular, Fryer et al. ( 2021 ) recently proposed a realistic “taxicab” payoff defined by a winner-takes-all dynamic that is in stark contrast to (2): Such nonlinear relations are prevalent in applications but fundamentally violate the additive model underpinning standard Shapley value…”
InfluentialGyeltshen, S., & Hayashi, K. (2025) · IEEE Asia-Pacific Conference on Geoscience, Electronics and Remote Sensing Technology (AGERS) · doi:10.1109/agers67633.2025.11446393
“Shapley Additive Explanations (SHAP) is grounded in cooperative game theory, originally developed by Lloyd Shapley to fairly distribute payoffs among players in coalition games [40].”
Correlations Matter in Explanations for Energy Systems
Nikoltchovska, A., Pütz, S., Götz, M., & Schäfer, B. (2026) · Künstliche Intell. · doi:10.1007/s13218-026-00904-4
“Fryer et al. [ 10] examined the theoretical foundations of Shapley values in machine learning contexts, highlighting potential pitfalls when assumptions are violated.”
Lv, P., Sun, L., & Liu, X. (2026) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2026.3681799 · Cited by 2
“In this study, we introduced a post-hoc interpretability analysis based on SHAP to further investigate the contribution of various features in the model and their interrelationships [53], [54], [55].”
Enhancing flood prediction through physics-driven typhoon feature engineering and machine learning
Zhang, Z., Xiao, Y., Chen, B., Long, K., Liang, F., & Liao, J. (2026) · PLoS ONE · doi:10.1371/journal.pone.0346237
“The additive assumption underlying SHAP may inadequately capture complex nonlinear feature interactions during extreme events [30].”
Guevara-Escobar, A., Lemus-Ramírez, V., García-Muñiz, J. G., Yabuta-Osorio, A. K., Vidales-Basurto, C. A., & Valdés-Aguirre, B. (2026) · Dairy · doi:10.3390/dairy7030031
Interpretable AI and Classical Statistics: Revisiting the Surrogate Model Paradigm
Abdelhay, O., Altamimi, T., Shatnawi, A., & Abdelhay, H. (2026) · International Conference on Advances in Artificial Intelligence and Machine Learning (AAIML) · doi:10.1109/aaiml67890.2026.11498135
“Recent studies and user experiences have shown that SHAP values can be potentially misleading when predictors are correlated, as they can be imprecise and even have the opposite sign [21-24].”
Explainable AI for HCI: a systematic review of a decade of studies on XAI tools
Alaqsam, A., & Sas, C. (2026) · Interacting with Computers · doi:10.1093/iwc/iwag022
Moussa, B., Ali, A. M., Abdeljaber, T., & Abdelbasset, M. M. (2026) · Journal of African Earth Sciences · doi:10.1016/j.jafrearsci.2026.106243 · Cited by 1
Kumar, N., Zhang, Y., Wiedemann, N., Oke, J. B., & Raubal, M. (2026) · Scientific Reports · doi:10.1038/s41598-026-58442-w
“In recent years, applying SHAP (Shapley Additive exPlanations) values has significantly advanced the investigation into model explainability, contributing to the development of explainable AI 55 .”
Less is More: XAI-Driven Feature Selection for Efficient Molecular Property Prediction
Durão, J. L., Dias, L. G., & Quiles, M. G. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-30491-9_14
A Preliminary Study on Explaining Risk of Code Changes using LLM-Based Prediction Models
Liu, Y., Jabre, K., Abreu, R., Carmichael, Z., Murali, V., Patel, A., Ge, J., Sun, W., Zhang, C., Mockus, A., et al. (2026) · ACM International Conference on AI-Powered Software (AIware) · doi:10.1145/3805760.3814916
Kwon, C. W., Kwon, M., Basith, S., Misra, S., Jang, Y. E., & Lee, G. (2026) · Protein Science · doi:10.1002/pro.70704 · Cited by 1
“Although the use of SHAP values in feature selection remains debated (Fryer et al. 2021; Wang et al. 2024), several studies have demonstrated their effectiveness (Marcílio and Eler 2020; Verhaeghe et al. 2022).”
Quantile-Based Trading on the German Continuous Intraday Market Using Weather Forecast Updates
Semmelmann, L., Miskiw, K. K., Thederan, M., & Weinhardt, C. (2026) · International Conference on the European Energy Market (EEM) · doi:10.1109/eem68581.2026.11589598
From Prediction to Insight: Explainable Ensemble Learning for CO2-to-Methanol Conversion
Nguyen, D., Guerrero‐Pérez, M. O., Rodrı́guez-Castellón, E., Nguyen, M. T., Trần, V. D., Truong, T. H., Nguyễn, X. P., & Hoang, A. T. (2026) · ACS Sustainable Chemistry & Engineering · doi:10.1021/acssuschemeng.6c03657
Afrash, M., Rahimi, M., Moulaei, K., Mostafazadeh, B., Rafsanjani, H., Evini, P. E. T., Parvin, M., Sharifi, M., salehi, S., Daeechini, A., et al. (2026) · International Journal of Emergency Medicine · doi:10.1186/s12245-026-01328-5
“The Shapley value has been a famous method for the determination of feature contribution to the model’s performance [43].”
Chen, L., Xue, W., Liu, Y., Wei, P., Lamont, B. B., & Yan, Z. (2026) · Plant Disease · doi:10.1094/pdis-05-25-1029-re
X-ZeroSec: Zero-Shot Evasion Attack Detection in Deep Learning-Based Beamforming Prediction With XAI
Senevirathna, T., Siniarski, B., Liyanage, M., & Wang, S. (2026) · IEEE Transactions on Machine Learning in Communications and Networking · doi:10.1109/tmlcn.2026.3724613
“Unlike prior works that employ XAI feature attribution merely as a feature selection mechanism [30], [31] (with which some work disagrees [32]), the novelty of our work lies in the use of TreeSHAP attributes as training and inference data to distill attack-discriminative signatures from a…”
Nguyen, D., Guerrero‐Pérez, M. O., Rodríguez-Castellón, E., Chuong, N. M., Hadiyanto, H., Le, V. H., Dong, V. H., Nguyễn, X. P., & Hoang, A. T. (2026) · iScience · doi:10.1016/j.isci.2026.117213
Multi-factor contributions to blast-induced rock fragmentation behavior
Li, X., He, C., Zhao, X., Xu, G., Wang, B., Liu, L., Ying, P., & Wang, X. (2026) · International Journal of Mining Science and Technology · doi:10.1016/j.ijmst.2026.08.003
Nashat, M., Zayed, T. M., Meguid, M. A., Ali, E., Arimiyawa, D., & Yang, J. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7451115
Association of Community Factors, Firearm Laws, and Pediatric Firearm Injuries
Diaz, X. N., Lin, A. L., Babcock, S. R., Salvi, A., Malveau, S. E., Mann, N. C., Goldstick, J. E., Carter, P. M., Cook, J. N. B., Song, X., et al. (2026) · PEDIATRICS · doi:10.1542/peds.2025-074149
An interpretable lithology identification model for small-sample well logging data
Mou, J., Wu, X., Ha, X., & Ding, J. (2026) · Journal of Applied Geophysics · doi:10.1016/j.jappgeo.2026.106533
Ma, T., Guo, Q., Yao, N., Gu, Z., Li, R., Zhou, J., Shi, L., & Su, J. (2026) · Renewable Energy · doi:10.1016/j.renene.2026.126458
Lim, J., Oh, H., Kim, M., & Shin, S. (2026) · International Journal of Energy Research · doi:10.1155/er/2674991
Xiaochang, L. I. U., Dong, W., Zhu, L., Zhang, X., Ao, X., 张学良, Wu, T., Chen, Z., & Qiao, R. (2026) · Applied Spatial Analysis and Policy · doi:10.1007/s12061-026-10038-7
“SHAP innovatively bridges Local Interpretable Model-agnostic Explanations (LIME) and Shapley values through an additive feature attribution approach (Fryer et al., 2021; Lundberg & Lee, 2017).”
Yan, Q., Bu, Y., Pan, Y., Jin, S., & Huang, W. (2026) · Satellite Navigation · doi:10.1186/s43020-026-00221-z
“SHAP provides a unified framework for quantifying the marginal contribution of each feature by considering all possible feature coalitions (Fryer et al., 2021).”
Metadata
- Authors
- Fryer, D., Strumke, I., & Nguyen, H.
- Affiliations
- The University of Queensland (AU)
- La Trobe University (AU)
- Venue
- IEEE Access
- Published
- 2021
- Volume
- 9
- Pages
- 144352-144360
- ISSN
- 2169-3536 (ISSN-L)
- OpenAlex
- W3131037858
- Semantic Scholar
- 6573fa74af238d1e3538c026997e31b9f67f19f7
- Cited by
- 347
- References
- 64
- Citations by year
- 2026: 56 · 2025: 102 · 2024: 86 · 2023: 58 · 2022: 34 · 2021: 11
- FWCI
- 16.63
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- IEEE Access (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Figshare (submittedVersion, cc-by)
- arXiv (Cornell University)
- La Trobe University (cc-by)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Game Theory and Voting Systems
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Keywords
- Shapley value, feature selection, explainable AI, axioms, SHAP, sage, counterexample, transferable utility games, attribution
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.