Beyond sigmoids: How to obtain well-calibrated probabilities from binary classifiers with beta calibration
Open accessThe DOI matches its Crossref record; 14 of the 16 references it cites were found in a registry or an index, and 2 are not checked yet.
Abstract
For optimal decision making under variable class distributions and misclassification costs a classifier needs to produce well-calibrated estimates of the posterior probability. Isotonic calibration is a powerful non-parametric method that is however prone to overfitting on smaller datasets; hence a parametric method based on the logistic sigmoidal curve is commonly used. While logistic calibration is designed for normally distributed per-class scores, we demonstrate experimentally that many classifiers including Naive Bayes and Adaboost suffer from a particular distortion where these score distributions are heavily skewed. In such cases logistic calibration can easily yield probability estimates that are worse than the original scores. Moreover, the logistic curve family does not include the identity function, and hence logistic calibration can easily uncalibrate a perfectly calibrated classifier. In this paper we solve all these problems with a richer class of parametric calibration maps based on the beta distribution. We derive the method from first principles and show that fitting it is as easy as fitting a logistic curve. Extensive experiments show that beta calibration is superior to logistic calibration for a wide range of classifiers: Naive Bayes, Adaboost, random forest, logistic regression, support vector machine and multi-layer perceptron. If the original classifier is already calibrated, then beta calibration learns a function close to the identity. On this we build a statistical test to recognise if the model deviates from being well-calibrated.
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 source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Beta calibration is equivalent to logistic calibration if 𝛽 1 = 𝛽 2 .8,9 Otherwise, it is more flexible than logistic calibration.”
From Calibrating machine learning approaches for probability estimation: A comprehensive comparison (2023) · cites this paper as [8]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“One approach that works well is to model the classwise scores with Beta distributions (Kull et al., 2017a(Kull et al., , 2017b)).”
From Classifier calibration: a survey on how to assess and improve predicted class probabilities (2023) · cites this paper as Kull et al., 2017b
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Isotonic calibration 8,12 is one of the most commonly used calibration approaches by the machine learning community.64”
From Calibrating machine learning approaches for probability estimation: A comprehensive comparison (2023) · cites this paper as [8]
The cited source says what the claim attributes to it.
In the citing paper
“One approach that works well is to model the classwise scores with Beta distributions (Kull et al., 2017a(Kull et al., , 2017b)).”
From Classifier calibration: a survey on how to assess and improve predicted class probabilities (2023) · cites this paper as Kull et al., 2017a
The cited source says what the claim attributes to it.
In the citing paper
“Beta calibration is specifically designed for probabilistic classifiers and alleviates both drawbacks of Platt scaling (Kull et al., 2017a, b).”
From Classifier calibration: a survey on how to assess and improve predicted class probabilities (2023) · cites this paper as Kull et al., 2017a, b
2 more claims about this paper
Create a free account to see every claim other papers make about this one.
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
Scikit-learn: Machine Learning in Python
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Müller, A., Nothman, J., Louppe, G., et al. (2012) · arXiv · doi:10.48550/arxiv.1201.0490 · Cited by 71,016 · cited 4×
“The second is the one implemented in Python’s Scikit-learn, based on Adaboost method SAMME (Zhu et al., 2009), referred to as Ada-S.”
MethodInfluentialThe WEKA data mining software: an update
Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P., & Witten, I. H. (2009) · ACM SIGKDD Explorations Newsletter · doi:10.1145/1656274.1656278 · Cited by 17,750 · cited 1×
“However, as the task is simply univariate logistic regression with feature s and label y , it can be solved using the standard logistic regression functionality in any machine learning toolkit, including WEKA (Hall et al., 2009) and Scikit-learn (Pedregosa et al., 2011).”
MethodFriedman, J., Hastie, T., & Tibshirani, R. (2000) · The Annals of Statistics · doi:10.1214/aos/1016218223 · Cited by 6,884
Predicting good probabilities with supervised learning
Niculescu-Mizil, A., & Caruana, R. (2005) · Proceedings of the 22nd international conference on Machine learning - ICML '05 · doi:10.1145/1102351.1102430 · Cited by 1,884 · cited 2×
“This was expected, since isotonic calibration is the only non-parametric method in this study and literature has shown that it needs more data than logistic calibration to fit a good calibration map (Niculescu-Mizil and Caruana, 2005).”
MethodHastie, T., Rosset, S., Zhu, J., & Zou, H. (2009) · Statistics and Its Interface · doi:10.4310/sii.2009.v2.n3.a8 · Cited by 1,875 · cited 1×
“The second is the one implemented in Python’s Scikit-learn, based on Adaboost method SAMME (Zhu et al., 2009), referred to as Ada-S.”
MethodTransforming classifier scores into accurate multiclass probability estimates
Zadrozny, B., & Elkan, C. (2002) · Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/775047.775151 · Cited by 1,280 · cited 3×
“The scores of a sufficiently calibrated classifier can be simply thresholded at a threshold directly derived from the misclassification cost (Zadrozny and Elkan, 2002).”
BackgroundMethodRobust classification for imprecise environments
Provost, F., & Fawcett, T. (2001) · Machine Learning · doi:10.1023/a:1007601015854 · Cited by 1,250
Platt, J. C. (2000) · Advances in Large-Margin Classifiers · doi:10.7551/mitpress/1113.003.0008 · Cited by 1,156 · cited 3×
“Logistic calibration was proposed by (Platt, 2000) to reduce overfitting by introducing a strong inductive bias which considers only calibration maps of the following form: where γ, δ are real-valued parameters with γ ≥ 0 to ensure that the calibration map is monotonically non-decreasing.”
MethodStatistical Comparisons of Classifiers over Multiple Data Sets
DemšarJanez (2006) · Journal of Machine Learning Research · doi:10.5555/1248547.1248548 · Cited by 1,074
Calibration of Probabilities: The State of the Art
Lichtenstein, S., Fischhoff, B., & Phillips, L. D. (1977) · Decision Making and Change in Human Affairs · doi:10.1007/978-94-010-1276-8_19 · Cited by 483
Cited by
Classifier calibration: a survey on how to assess and improve predicted class probabilities
Silva Filho, T., Song, H., Perello-Nieto, M., Santos-Rodriguez, R., Kull, M., & Flach, P. (2023) · Machine Learning · doi:10.1007/s10994-023-06336-7 · Cited by 228
“( ) Similar cross-validation training is also adopted as proposed in Kull et al. (2017a).”
BackgroundMethodInfluentialSelf-citationKull, M., Perello-Nieto, M., Kängsepp, M., Filho, T. S., Song, H., & Flach, P. (2019) · arXiv · doi:10.48550/arxiv.1910.12656 · Cited by 206
“For 2 classes, the Dirichlet calibration map family coincides with the beta calibration map family [15].”
MethodSelf-citationVerified Uncertainty Calibration
Kumar, A., Liang, P., & Ma, T. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.10155 · Cited by 131
“We first describe the scaling family we use, which is Platt scaling after applying a log-transform [12], otherwise known as beta calibration [45].”
MethodPerformance Evaluation in Machine Learning: The Good, the Bad, the Ugly, and the Way Forward
Flach, P. A. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.33019808 · Cited by 221
Self-citationIdentifying and prioritizing potential human-infecting viruses from their genome sequences
Mollentze, N., Babayan, S. A., & Streicker, D. G. (2021) · PLoS Biology · doi:10.1371/journal.pbio.3001390 · Cited by 138
Assistant Project Manager, Ahmmed, M. J., Kamal, M., & VP & Unit Head, I. (. (2023) · American Journal of Interdisciplinary Studies · doi:10.63125/60amyk26 · Cited by 128
Uncertainty quantification in drug design
Mervin, L., Johansson, S., Semenova, E., Giblin, K. A., & Engkvist, O. (2020) · Drug Discovery Today · doi:10.1016/j.drudis.2020.11.027 · Cited by 108
Local Temperature Scaling for Probability Calibration
Ding, Z., Han, X., Liu, P., & Niethammer, M. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00681 · Cited by 148
“Further, beta calibration [35] extends logistic calibration to overcome the situation where per-class score distributions are heavily skewed.”
BackgroundMix-n-match: Ensemble and compositional methods for uncertainty calibration in deep learning
Zhang, J., Kailkhura, B., & Han, Y.-J. (2020) · Unpublished · doi:10.13140/rg.2.2.11390.13125 · Cited by 86
A general approach for determining applicability domain of machine learning models
Schultz, L. E., Wang, Y., Jacobs, R. M., & Morgan, D. D. (2025) · npj Computational Materials · doi:10.1038/s41524-025-01573-x · Cited by 87
“Uncertainty calibration has been extensively studied in both classi fi cation and regression settings 21 – 25 .”
Iacucci, M., Santacroce, G., Zammarchi, I., Maeda, Y., Del Amor, R., Meseguer, P., Kolawole, B. B., Chaudhari, U., Di Sabatino, A., Danese, S., et al. (2024) · The Lancet Gastroenterology & Hepatology · doi:10.1016/s2468-1253(24)00053-0 · Cited by 73
Calibrating machine learning approaches for probability estimation: A comprehensive comparison
Ojeda, F. M., Jansen, M. L., Thiéry, A., Blankenberg, S., Weimar, C., Schmid, M., & Ziegler, A. (2023) · Statistics in Medicine · doi:10.1002/sim.9921 · Cited by 71
“8,15 However, a comprehensive comparison of the various calibration approaches on different learning machines for probability estimation using Monte-Carlo simulation studies and real data sets with training and temporal and/or external validation data is lacking.”
BackgroundMethodInfluentialFSL-Kla: A few-shot learning-based multi-feature hybrid system for lactylation site prediction
Jiang, P., Ning, W., Shi, Y., Liu, C., Mo, S., Zhou, H., Liu, K., & Guo, Y. (2021) · Computational and Structural Biotechnology Journal · doi:10.1016/j.csbj.2021.08.013 · Cited by 69
“In general, the sigmoid method was effective when the un-calibrated model was under-confident and had similar calibration errors for both high and low outputs [55].”
MethodFraud Detection Using Neural Networks: A Case Study of Income Tax
Murorunkwere, B. F., Tuyishimire, O., Haughton, D., & Nzabanita, J. (2022) · Future Internet · doi:10.3390/fi14060168 · Cited by 59
“…a loss function that is well suited for binary classification tasks, as it compares predicted probabilities to actual classes by measuring the distance between the prediction and actual output when training and the computed log-loss shows how close or far the prediction is to the actual values [32].”
BackgroundHumphrey, A., Kuberski, W., Białek, J., Perrakis, N., Cools, W., Nuyttens, N., Elakhrass, H., & Cunha, P. A. C. (2022) · Monthly Notices of the Royal Astronomical Society Letters · doi:10.1093/mnrasl/slac120 · Cited by 69
Artificial Intelligence in Medicine: Text Mining of Health Care Workers’ Opinions
Nitiéma, P. (2023) · Journal of Medical Internet Research · doi:10.2196/41138 · Cited by 47
Rahman, M. A., & Haque, B. M. T. (2022) · doi:10.63125/9hd4e011 · Cited by 42
Clairvoyant: AdaBoost with Cost-Enabled Cost-Sensitive Classifier for Customer Churn Prediction
Thakkar, H. K., Desai, A. B., Ghosh, S. K., Singh, P., & Sharma, G. (2022) · Computational Intelligence and Neuroscience · doi:10.1155/2022/9028580 · Cited by 41
“*is normalization of probability calibration [37] is mandatory to execute the weight update operation in the reweighting equation of the AdaBoostWithCost algorithm as the weight (probability) distribution of each data point varies between 0 and 1.”
MethodDong, J., Jiang, Z., Pan, D., Chen, Z., Guan, Q., Zhang, H., Gui, G., & Gui, W. (2025) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2025.3565159 · Cited by 45
Distribution Calibration for RegressionPreprint
Song, H., Diethe, T., Kull, M., & Flach, P. A. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.06023 · Cited by 37
“However, from simple models (e.g. Naı̈ve Bayes) to complex ones (e.g. (deep) Neural Networks (NNs)), poor calibration is often observed, irrespective of the model’s complexity or its probabilistic nature [9, 16].”
MethodInfluentialSelf-citationSeries Arc Fault Identification in DC Distribution Based on Random Forest Predicted Probability
Le, V., Miller, C., Tsao, B., & Yao, X. (2022) · IEEE Journal of Emerging and Selected Topics in Power Electronics · doi:10.1109/jestpe.2022.3228421 · Cited by 36
“It is suggested that the Sigmoid function works best if the calibration error is symmetrical, where the algorithm outputs each class normally distributed with the same variance [53].”
BackgroundDeniffel, D., Abraham, N., Namdar, K., Dong, X., Salinas, E., Milot, L. M., Khalvati, F., & Haider, M. A. (2020) · European Radiology · doi:10.1007/s00330-020-07030-1 · Cited by 38
Handling imbalanced data for aircraft predictive maintenance using the BACHE algorithm
Dangut, M. D., Skaf, Z., & Jennions, I. K. (2022) · Applied Soft Computing · doi:10.1016/j.asoc.2022.108924 · Cited by 37
Probability Calibration for Knowledge Graph Embedding ModelsPreprint
Tabacof, P., & Costabello, L. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.10000 · Cited by 36
“Future work will evaluate additional calibration algorithms, such as beta calibration (Kull et al., 2017) or Bayesian binning (Naeini et al., 2015).”
MethodMervin, L., Afzal, A. M., Engkvist, O., & Bender, A. (2020) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.0c00476 · Cited by 30
He, C., Wang, R., & Chen, X. (2021) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw53098.2021.00395 · Cited by 35
“For more recent probability calibration methods, the readers could refer to [35, 12, 34].”
BackgroundStaartjes, V. E., & Kernbach, J. M. (2020) · Journal of Neurosurgery Spine · doi:10.3171/2019.12.spine191503 · Cited by 29
Non-Parametric Calibration for Classification
Wenger, J., Kjellström, H., & Triebel, R. (2019) · elib (German Aerospace Center) · doi:10.48550/arxiv.1906.04933 · Cited by 114
“More recently, Bayesian binning into quantiles (Naeini et al., 2015) and beta calibration (Kull et al., 2017a) for binary classification and temperature scaling (Guo et al., 2017) for multi-class problems were proposed.”
MethodOn (assessing) the fairness of risk score models
Petersen, E., Ganz, M., Holm, S., & Feragen, A. (2023) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3593013.3594045 · Cited by 36
“We perform randomized hyperparameter search using five-fold cross-validation on the training set and calibrate the resultingmodel using beta calibration [52] (first case study, because of relatively”
MethodQuantifying the Confidence of Anomaly Detectors in Their Example-Wise Predictions
Perini, L., Vercruyssen, V., & Davis, J. J. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-67664-3_14 · Cited by 35
Zavagno, A., Dupé, F.-X., Bensaid, S., Schisano, E., Causi, G. L., Gray, M., Molinari, S., Elia, D., Lambert, J.-C., Brescia, M., et al. (2022) · Astronomy and Astrophysics · doi:10.1051/0004-6361/202244103 · Cited by 27
Distribution-free binary classification: prediction sets, confidence intervals and calibrationPreprint
Gupta, C., Podkopaev, A., & Ramdas, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.10564 · Cited by 106
Image processing techniques to identify tomato quality under market conditions
Abekoon, T., Sajindra, H., Jayakody, A., Samarakoon, E. R. J., & Rathnayake, U. (2024) · Smart Agricultural Technology · doi:10.1016/j.atech.2024.100433 · Cited by 25
Post hoc calibration of medical segmentation models
Rousseau, A.-J., Becker, T., Appeltans, S., Blaschko, M. B., & Valkenborg, D. (2025) · Discover Applied Sciences · doi:10.1007/s42452-025-06587-0 · Cited by 28
“This method was implemented as a bivariate logistic with three parameters in the manner described in [30].”
Ansah-Narh, T., Nortey, E. N. N., Proven-Adzri, E., & Opoku-Sarkodie, R. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.125133 · Cited by 19
Confidence Intervals for Class Prevalences under Prior Probability Shift
Tasche, D. (2019) · Machine Learning and Knowledge Extraction · doi:10.3390/make1030047 · Cited by 21
“Note, however, the promising recent progress made on this issue (Kull et al., 2017).”
BackgroundA fault detector/classifier for closed-ring power generators using machine learning
Quintanilha, I. M., Elias, V. R. M., da Silva, F. B., Fonini, P., da Silva, E. A. B., Netto, S. L., Apolinário, J. A., de Campos, M. L. R., Martins, W. A., Wold, L. E., et al. (2021) · Reliability Engineering & System Safety · doi:10.1016/j.ress.2021.107614 · Cited by 19
“…the gap between the probability levels associated with the normal and anomalous operating modes by distorting the fault probability 𝑝 through a sigmoid function, such that the new probability 𝑝 ′ is with 𝐴 and 𝐵 obtained by minimizing the log-likelihood function for the training dataset [67].”
BackgroundTop-label calibration and multiclass-to-binary reductionsPreprint
Gupta, C., & Ramdas, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.08353 · Cited by 25
“Thus, there is interest in studying statistically feasible relaxations of canonical notion, such as confidence calibration [Guo et al., 2017] and class-wise calibration [Kull et al., 2017]. ar X iv :2 10 7.”
BackgroundPredictive analytics for cardiovascular patient readmission and mortality: An explainable approach
Huberts, L. C. E., Li, S., Blake, V., Jorm, L. R., Yu, J. S., Ooi, S., & Gallego, B. (2024) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2024.108321 · Cited by 17
Li, T., Xu, K., Krishnan, A. R., Gao, R., Kammer, M. N., Antic, S., Xiao, D., Knight, M. A., Martinez, Y., Paez, R., et al. (2025) · Radiology Artificial Intelligence · doi:10.1148/ryai.230506 · Cited by 16
Stoppa, F., Bhattacharyya, S., de Austri, R. R., Vreeswijk, P. M., Caron, S., Zaharijaš, G., Bloemen, S., Principe, G., Malyshev, D. V., Vodeb, V., et al. (2023) · Astronomy and Astrophysics · doi:10.1051/0004-6361/202347576 · Cited by 19
Distribution-free calibration guarantees for histogram binning without sample splittingPreprint
Gupta, C., & Ramdas, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2105.04656 · Cited by 63
“A number of non-binned methods for calibrating neural networks have displayed good performance on some tasks [Guo et al., 2017, Kull et al., 2017, Lakshminarayanan et al., 2017].”
A Comprehensive Review of Classifier Probability Calibration Metrics
Lane, R. O. (2026) · Data Science · doi:10.1177/24518492261436231 · Cited by 18
Röchner, P., Marques, H. O., Campello, R. J. G. B., & Zimek, A. (2024) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-024-01056-5 · Cited by 15
“The evaluation of probability estimates in supervised learning has been studied extensively (Platt et al. 1999; Niculescu-Mizil and Caruana 2005; Kull et al. 2017).”
Rethinking and recomputing the value of machine learning models
Sayin, B., Yang, J., Chen, X., Passerini, A., & Casati, F. (2025) · Artificial Intelligence Review · doi:10.1007/s10462-025-11242-6 · Cited by 15
Tasche, D. (2021) · Statistics · doi:10.1080/02331888.2021.2016767 · Cited by 12
“Is then everything lost if no sufficient scoring classifier can be identified? Without referring to the notions of sufficiency or bipartite ranking, Langford and Zadrozny (2005) presented ‘probing reduction’ as a way to ‘reduce learning an estimator of class probability membership [sic] to learning binary classifiers’.”
BackgroundInfluentialDetecting Concept Drift In Medical Triage
Huggard, H., Koh, Y. S., Dobbie, G., & Zhang, E. (2020) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3397271.3401228 · Cited by 12
“CDDM achieves this by detecting changes in the calibration [5] of the base learner, rather than changes in accuracy.”
BackgroundMulti-Loss Sub-Ensembles for Accurate Classification with Uncertainty EstimationPreprint
Achrack, O., Kellerman, R., & Barzilay, O. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.01917 · Cited by 12
“Post-Hoc Calibration [26] aims at learning a re-calibration matrix for reducing the ECE on the predictions produced by the softmax activation [27].”
MethodDeep Probability EstimationPreprint
Liu, S., Kaku, A., Zhu, W., Leibovich, M., Mohan, S., Yu, B., Huang, H., Zanna, L., Razavian, N., Weed, J., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.10734 · Cited by 13
“• Beta/Dirchlet calibration (Dir-ODIR) (Kull et al., 2017; 2019) assumes that the probabilities can be parametrized by a Beta/Dirchlet distribution i.”
BackgroundMethodInfluentialMachine learning models for temporally precise lapse prediction in alcohol use disorder.
Wyant, K., Sant'Ana, S. J. K., Fronk, G. E., & Curtin, J. J. (2024) · Journal of Psychopathology and Clinical Science · doi:10.1037/abn0000901 · Cited by 11
“We calibrated predicted probabilities using the beta distribution to support optimal decision-making under variable outcome distributions (Kull et al., 2017).”
MethodBallinari, D., & Bearth, N. (2026) · Journal of Applied Econometrics · doi:10.1002/jae.70057 · Cited by 11
“Beta scaling can be shown to be equivalent to a logistic regression on log p ˆ p p x qq and log p 1 ´ ˆ p p x qq (Kull et al., 2017).”
InfluentialTemporal Probability CalibrationPreprint
Leathart, T., & Polaczuk, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.02644 · Cited by 10
“Beta scaling (Kull et al., 2017) and Dirichlet scaling (Kull et al., 2019) are, in a practical sense, very similar to Platt scaling and matrix scaling respectively.”
BackgroundResultMapping HDX-MS Data to Protein Conformations through Training Ensemble-Based Models
Salmas, R. E., Harris, M. J., & Borysik, A. J. (2023) · Journal of the American Society for Mass Spectrometry · doi:10.1021/jasms.3c00145 · Cited by 9
Model Uncertainty Quantification: A Post Hoc Calibration Approach for Heart Disease Prediction
Odesola, P. A., Adegoke, A. A., & Babalola, I. (2025) · Journal of Engineering Research and Sciences · doi:10.55708/js0412003 · Cited by 9
Selection of powerful radio galaxies with machine learning
Carvajal, R., Matute, I., Afonso, J., Norris, R. P., Luken, K., Sanchéz-Sáez, P., Cunha, P. A. C., Humphrey, A., Messias, H., Amarantidis, S., et al. (2023) · Astronomy and Astrophysics · doi:10.1051/0004-6361/202245770 · Cited by 11
Calibrating Deep Neural Network Classifiers on Out-of-Distribution DatasetsPreprint
Shao, Z., Yang, J., & Ren, S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.08914 · Cited by 13
Trajdos, P., & Burduk, R. (2024) · Knowledge-Based Systems · doi:10.1016/j.knosys.2024.112411 · Cited by 9
A plug-in approach to maximising precision at the top and recall at the topPreprint
Tasche, D. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1804.03077 · Cited by 8
“• Given recent progress in the estimation of posterior probabilities (see Kull et al., 2017, and the references therein), a plug-in approach based on thresholding the posterior positive class probability might turn out to have a competitive edge against other approaches.”
BackgroundMethodH-Calibration: Rethinking Classifier Recalibration With Probabilistic Error-Bounded Objective
Huang, W., Cao, G., Xia, J., Chen, J., Wang, H., & Zhang, J. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3582796 · Cited by 8
“Furthermore, many calibrators that were initially proposed for binary classifier [11], [123], [124], [125], [127], [129], [131], [140], when extended to multiclass calibration using strategies like one-vs-rest [67], also face this problem.”
InfluentialCombining Priors with Experience: Confidence Calibration Based on Binomial Process Modeling
Dong, J., Jiang, Z., Pan, D., & Yu, H. (2025) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v39i15.33792 · Cited by 9
Deniffel, D., Perlis, N., Ghai, S., Salinas-Miranda, E., Namdar, K., Klotz, L. H., Zlotta, A. R., Finelli, A., & Haider, M. A. (2025) · World Journal of Urology · doi:10.1007/s00345-025-05649-7 · Cited by 6
“To ensure accurate risk predictions for the target population of men with PI-RADS 3 index lesions, we applied a recalibration step using beta-calibration [22] to align the risk distribution with the prevalence observed in the PI-RADS 3 subgroup of the training cohort.”
Yego, F., Kabano, I. H., Murorunkwere, B. F., & Nzabanita, J. (2021) · SN Business & Economics · doi:10.1007/s43546-021-00104-2 · Cited by 7
Nomqupu, S., Sali, A., Nyamugama, A., & Ndou, N. N. (2024) · Applied Sciences · doi:10.3390/app14072670 · Cited by 6
“The sigmoid calibration function can calibrate potholes by relying on two parameters, namely (a) a location parameter ( m ), which specifies the sigmoidal mid-point, where the calibrated score is 0.5, and (b) a shape parameter ( γ ), which specifies the slope of the sigmoid at the midpoint [34].”
MethodInfluentialEstimating Uncertainty Intervals from Collaborating Networks
Zhou, T. H., Li, Y., Wu, Y. S., & Carlson, D. (2020) · PubMed · doi:10.48550/arxiv.2002.05212 · Cited by 6
“In the classical logistic regression setting, the probability is usually well calibrated (Kull et al., 2017).”
BackgroundOh, N., Oh, T. Y.-C., Hsu, J., Kim, D. W., Yu, B., Cho, J., Seok, J., Lee, J. Y., Jang, W., Kim, J., et al. (2026) · Nature Communications · doi:10.1038/s41467-025-68198-y · Cited by 6
“We applied a posthoc calibration procedureusing the beta calibration method based on the beta distribution 35 .”
A Model-Agnostic Feature Selection Technique to Improve the Performance of One-Class Classifiers
Hancock, J., Bauder, R. A., & Khoshgoftaar, T. M. (2023) · IEEE International Conference on Tools with Artificial Intelligence (ICTAI) · doi:10.1109/ictai59109.2023.00021 · Cited by 5
“Therefore, we applied sigmoid calibration to convert the GMM outputs into well-calibrated probabilities [22], [23].”
MethodMiraz, A. I. (2025) · American Journal of Interdisciplinary Studies · doi:10.63125/869qdj55 · Cited by 11
From Uncertainty to Precision: Enhancing Binary Classifier Performance through CalibrationPreprint
Machado, A. F., Arthur Charpentier, A., Flachaire, E., Gallic, E., & Hu, F. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2402.07790 · Cited by 5
“When using scores generated by a model predicting the probability of a binary event, the literature advocates recalibrating the model by applying a transformation to the scores (Platt, 1999; Zadrozny and Elkan, 2002; Kull et al., 2017).”
BackgroundMethodInfluentialOUEDRAOGO, D. N. (2021) · DOAJ (DOAJ: Directory of Open Access Journals) · Cited by 5
The art of transfer learning: An adaptive and robust pipeline
Wang, B., Wu, Y., & Ye, C. (2023) · Stat · doi:10.1002/sta4.582 · Cited by 4
“Other examples of classification calibration methods include beta calibration (Kull et al., 2017), isotonic regression (Zadrozny & Elkan, 2002), and confidence calibration (Guo et al., 2017), among many others.”
MethodOn the Dark Side of Calibration for Modern Neural Networks.Preprint
Singh, A. K., Bay, A., Sengupta, B., & Mirabile, A. (2021) · arXiv (Cornell University) · Cited by 7
On Deep Neural Network Calibration by Regularization and its Impact on RefinementPreprint
Singh, A. K., Bay, A., Sengupta, B., & Mirabile, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.09385 · Cited by 4
“Some of the recently proposed approaches are temperature scaling (Guo et al. 2017), scaling and binning calibration (Kumar et al. 2019), Dirichlet calibration (Kull et al. 2019), and beta calibration (Kull et al. 2017).”
BackgroundKomninos, P., & Zarouchas, D. S. (2025) · Array · doi:10.1016/j.array.2025.100454 · Cited by 3
The development of fragility curves using calibrated probabilistic classifiers
Saleh, E. F. (2024) · Structures · doi:10.1016/j.istruc.2024.106618 · Cited by 3
Local variational Probabilistic Minimax Active Learning
Ghafarian, S. H. (2022) · Expert Systems with Applications · doi:10.1016/j.eswa.2022.118538 · Cited by 3
Gerevini, A., Maroldi, R., Olivato, M., Putelli, L., & Serina, I. (2023) · IEEE Access · doi:10.1109/access.2023.3296260 · Cited by 3
“Since standard approaches such as calibration [11] or NGBoost [12] cannot be effectively adopted in our context, mainly due to the limited size of the datasets [13], [14], [15], we designed a new algorithm to identify and rule out uncertain predictions.”
BackgroundImpact of pectoral muscle removal on deep-learning-based breast cancer risk prediction
Klaneček, Ž., Wang, Y., Wagner, T., Cockmartin, L., Marshall, N., Schott, B., Deatsch, A. E., Studen, A., Jarm, K., Krajc, M., et al. (2025) · Physics in Medicine and Biology · doi:10.1088/1361-6560/adb367 · Cited by 3
“All models were calibrated with Beta calibration (Kull et al 2017b), where the uncalibrated probabilities ( z ) are transformed to actual probabilities p by training a function with three tunable parameters α,β,γ : ) .”
InfluentialStoppa, F., Bhattacharyya, S., de Austri, R. R., Vreeswijk, P. M., Caron, S., Zaharijaš, G., Bloemen, S., Principe, G., Malyshev, D. V., Vodeb, V., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2307.14456 · Cited by 4
Supervised Learning Approaches for Robust Predictive Modelling in Data Science
Dond, V. T. (2025) · International Journal for Research in Applied Science and Engineering Technology · doi:10.22214/ijraset.2025.73756 · Cited by 2
A Closer Look at Probability Calibration of Knowledge Graph Embedding
Zhu, R., Wang, F., Bundy, A., Li, X., Nuamah, K., Xu, L., Mauceri, S., & Pan, J. Z. (2022) · International Joint Conference on Knowledge Graphs (IJCKG) · doi:10.1145/3579051.3579072 · Cited by 2
“…are desired, especially in high-stake decision-making tasks, like medical diagnosis, autonomous driving, etc. Uncalibrated probabilistic models will lead to under-estimated or over-estimated risks [11, 38], while calibrated probabilities are necessary to make optical decisions [15, 45].”
BackgroundValue-Based Hybrid Intelligence
Sayin, B., Yang, J., Passerini, A., & Casati, F. (2023) · Frontiers in artificial intelligence and applications · doi:10.3233/faia230100 · Cited by 2
Molladavoudi, S., & Yung, W. (2023) · AStA Wirtschafts- und Sozialstatistisches Archiv · doi:10.1007/s11943-023-00331-z · Cited by 4
“There are various ways to adjust the predicted probabilities to match the empirical observations, for instance Platt scaling Platt (1999), Beta and Dirichlet calibrations Kull et al. (2017), isotonic regression Zadrozny and Elkan (2001), etc.”
Qu, C., Luna, A. J., Li, T. Z., Zhu, J., Guo, J., Xiong, J., Sandler, K. L., Landman, B. A., & Huo, Y. (2026) · SPIE Medical Imaging: Computer-Aided Diagnosis · doi:10.1117/12.3087567 · Cited by 3
Class-wise and reduced calibration methods
Panchenko, M., Benmerzoug, A., & Delgado, M. D. B. (2022) · International Conference on Machine Learning and Applications · doi:10.1109/icmla55696.2022.00180 · Cited by 5
“In the binary setting, [9] use Beta distributions and show them to outperform logistic models in many situations, whereas [10] evaluate multiple methods for scoring of loan applications and find non-parametric methods to outperform parametric ones.”
BackgroundSHAP as a Data Reduction Technique for Highly Imbalanced Big Data
Hancock, J., Bauder, R. A., & Khoshgoftaar, T. M. (2025) · International Journal of Artificial Intelligence Tools · doi:10.1142/s0218213025400019 · Cited by 1
Enhancing Synthetic Generated-Images Detection through Post-Hoc Calibration
Dimitri, G. M., Tondi, B., & Barni, M. (2025) · IEEE/CVF Winter Conference on Applications of Computer Vision Workshops (WACVW) · doi:10.1109/wacvw65960.2025.00087 · Cited by 1
“…in fact, we assume that the ratio of the score distributions, for both binary classes, are similar to the ratio of two Gaussians distributions with the same variances, while in the beta regression case, it is assumed that they can be approximated as the ratio of two beta distributions [15].”
InfluentialConsistently Evaluating Record Linkage Classifiers
Christen, P., Ziyad, S., & Nanayakkara, C. (2026) · Journal of Data and Information Quality · doi:10.1145/3786768 · Cited by 1
“5 We then trained these classifiers using ten-fold cross validation, and calibrated the models to obtain classification scores ( 𝑠 𝑖𝑗 ) that are estimated probabilities for a record pair to be a match [24].”
WWAggr: A Window Wasserstein-Based Aggregation for Ensemble Change Point Detection
Stepikin, A., Romanenkova, E., & Zaytsev, A. (2025) · Proceedings · doi:10.1109/icdm65498.2025.00163 · Cited by 1
“Further, we operate in the space of the predicted CP scores, detecting distributional shifts there.”
Enhancing morpho-kinematic analysis for plant water stress classification through leaf movements
Polilli, W., Antonini, A., Platani, C., Stagnari, F., & Galieni, A. (2026) · Smart Agricultural Technology · doi:10.1016/j.atech.2026.101930 · Cited by 1
Bigan, E., & Dufour, S. (2026) · JMIR AI · doi:10.2196/83853 · Cited by 1
“Various post hoc calibration methods were applied to raw LLM probabilities and compared using the Brier score loss: isotonic calibration [21], Platt calibration [22], and beta calibration [23].”
Sayin, B., Yang, J., Passerini, A., & Casati, F. (2023) · Frontiers in artificial intelligence and applications · doi:10.3233/faia230085 · Cited by 1
Dong, J., Jiang, Z., Pan, D., Yu, H., Guan, Q., Ding, B., & Gui, W. (2023) · SSRN Electronic Journal · doi:10.2139/ssrn.4677185 · Cited by 1
A Monte Carlo Framework for Calibrated Uncertainty Estimation in Sequence PredictionPreprint
Yang, Q., Zhu, W., Keslin, J., Zanna, L., Rudner, T. G. J., & Fernandez‐Granda, C. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2410.23272 · Cited by 2
“Various methods have been developed to enhance calibration in classification, including post-processing the logits [Gupta et al., 2021, Kull et al., 2017, 2019], ensem-bling methods [Zhang et al., 2020a, Lakshminarayanan et al., 2017, Maddox et al., 2019], soft labeling [Mukhoti et al., 2020,…”
MethodHuang, J., Narasimha, S. M., Patel, A. N., Sristi, R. D., Mishne, G., & Gilja, V. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.04.02.715749 · Cited by 2
“This gap between accuracy and reliability has motivated extensive work on uncertainty quantification, including post-hoc calibration methods [64–66], conformal prediction [67,68], Bayesian neural networks [69], and ensemble approaches [70,71].”
Optimizing Marketing Campaign Targeting Using Uncertainty-Based Predictive Modelling
Gunnarsson, B. R., Broucke, S. V., & Weerdt, J. D. (2019) · IEEE ... International Conference on Data Mining workshops · doi:10.1109/icdmw.2019.00055 · Cited by 2
“To do so, we assume the probability distribution of the classifier is well calibrated [16].”
BackgroundProbability-driven scoring functions in combining linear classifiers
Trajdos, P., & Burduk, R. (2022) · JUCS - Journal of Universal Computer Science · doi:10.3897/jucs.80747
“In this task, we want to provide an estimation of the class posterior probability distribution [44].”
MethodEvolutionary Optimization of High-Coverage Budgeted ClassifiersPreprint
Hamilton, N. H., & Fulp, E. W. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.13067
“We also assume that these methods are sufficiently well-calibrated (Kull et al., 2017) such that their confidence scores roughly approximate true class probabilities as in logistic regression —the classification algorithm utilized by EMSCO at each stage.”
MethodCan Ensemble Calibrated Learning Enhance Link Prediction? A Study on Commonsense Knowledge
Racharak, T., Jearanaiwongkul, W., & Thwe, K. M. (2023) · Lecture notes in computer science · doi:10.1007/978-981-99-5837-5_16
Data Reduction to Improve the Performance of One-Class Classifiers on Highly Imbalanced Big Data
Hancock, J., & Khoshgoftaar, T. M. (2023) · International Conference on Machine Learning and Applications · doi:10.1109/icmla58977.2023.00071
“To address this limitation, we applied sigmoid calibration to convert the raw GMM outputs into probability values that are well-calibrated across their entire range [24], [25], [26].”
MethodTowards Calibrated, Sharp, and Interpretable Probabilistic Prediction
Sun, Z. (2024) · Deep Blue (University of Michigan) · doi:10.7302/24993 · Cited by 1
Metadata
- Authors
- Kull, M., Silva Filho, T. M., & Flach, P.
- Affiliations
- University of Tartu (EE)
- University of Bristol (GB)
- Venue
- Electronic Journal of Statistics
- Publisher
- Institute of Mathematical Statistics
- Published
- 2017-01-01
- Volume
- 11
- Issue
- 2
- Pages
- 5052-5080
- ISSN
- 1935-7524 (ISSN-L)
- OpenAlex
- W2773235741
- Cited by
- 111
- References
- 15
- Citations by year
- 2026: 22 · 2025: 16 · 2024: 11 · 2023: 16 · 2022: 10 · 2021: 17 · 2020: 11 · 2019: 7 · 2018: 1
- FWCI
- 3.10
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- Locations
- Electronic Journal of Statistics (publishedVersion, cc-by)
- Explore Bristol Research (submittedVersion, cc-by)
- Project Euclid (Cornell University) (submittedVersion)
- Bristol Research (University of Bristol) (submittedVersion, cc-by)
- Funders
- Eesti Teadusagentuur (PUT1458)
- Conselho Nacional de Desenvolvimento Científico e Tecnológico
- Engineering and Physical Sciences Research Council (EP/K031910/1, EP/K018728/1)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Imbalanced Data Classification Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Data Classification
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Keywords
- probability calibration, binary classifiers, posterior probability estimation, isotonic calibration, naive Bayes, AdaBoost, random forest, support vector machine, multilayer perceptron, calibration tests
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.