Adjusting the Outputs of a Classifier to New a Priori Probabilities: A Simple Procedure
Open accessThe DOI matches its Crossref record; 26 of the 26 references it cites were found in a registry or an index.
Abstract
It sometimes happens (for instance in case control studies) that a classifier is trained on a data set that does not reflect the true a priori probabilities of the target classes on real-world data. This may have a negative effect on the classification accuracy obtained on the real-world data set, especially when the classifier's decisions are based on the a posteriori probabilities of class membership. Indeed, in this case, the trained classifier provides estimates of the a posteriori probabilities that are not valid for this real-world data set (they rely on the a priori probabilities of the training set). Applying the classifier as is (without correcting its outputs with respect to these new conditions) on this new data set may thus be suboptimal. In this note, we present a simple iterative procedure for adjusting the outputs of the trained classifier with respect to these new a priori probabilities without having to refit the model, even when these probabilities are not known in advance. As a by-product, estimates of the new a priori probabilities are also obtained. This iterative algorithm is a straightforward instance of the expectation-maximization (EM) algorithm and is shown to maximize the likelihood of the new data. Thereafter, we discuss a statistical test that can be applied to decide if the a priori class probabilities have changed from the training set to the real-world data. The procedure is illustrated on different classification problems involving a multilayer neural network, and comparisons with a standard procedure for a priori probability estimation are provided. Our original method, based on the EM algorithm, is shown to be superior to the standard one for a priori probability estimation. Experimental results also indicate that the classifier with adjusted outputs always performs better than the original one in terms of classification accuracy, when the a priori probability conditions differ from the training set to the real-world data. The gain in classification accuracy can be significant.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
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?
In the citing paper
“However, if wellcalibrated fraud probabilities are needed, then the posterior probabilities can be adjusted (Saerens et al., 2002).”
From Data engineering for fraud detection (2021) · cites this paper as Saerens et al., 2002
Citations
Cites
Maximum Likelihood from Incomplete Data Via the EM Algorithm
Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977) · Journal of the Royal Statistical Society Series B: Statistical Methodology · doi:10.1111/j.2517-6161.1977.tb01600.x · Cited by 50,444
Probability, Random Variables, and Stochastic Processes.
Abrahams, J., & Papoulis, A. (1984) · Journal of the American Statistical Association · doi:10.2307/2288754 · Cited by 16,226
UCI Repository of machine learning databases
Blake, C. L. (1998) · Medical Entomology and Zoology · Cited by 10,582
The EM Algorithm and Extensions
Kushary, D. (1998) · Technometrics · doi:10.1080/00401706.1998.10485534 · Cited by 5,417
An Introduction to the Theory of Statistics
(1911) · Nature · doi:10.1038/087345a0 · Cited by 4,394
Probability, Random Variables, and Stochastic Processes
Papoulis, A., & Hoffman, J. G. (1967) · Physics Today · doi:10.1063/1.3034123 · Cited by 4,107
Journal of the Royal Statistical Society (B)
Fildes, R. (1993) · International Journal of Forecasting · doi:10.1016/0169-2070(93)90088-5 · Cited by 3,437
Text classification from labeled and unlabeled documents using EM
Nigam, K., Mccallum, A. K., Thrun, S., & Mitchell, T. (2000) · Machine Learning · doi:10.1023/a:1007692713085 · Cited by 2,736 · cited 1×
“For instance, Nigam et al. (2000) use the EM algorithm to learn classiers that take advantage of both labeled and unlabeled data.”
MethodDiscriminant Analysis and Statistical Pattern Recognition.
Hawkins, D. M., & McLachlan, G. J. (1993) · Journal of the American Statistical Association · doi:10.2307/2290355 · Cited by 2,633
Mixture models : inference and applications to clustering
McLachlan, G. J., & Basford, K. E. (1988) · M. Dekker eBooks · Cited by 2,094
Cited by
Items 101–200 of 362
Fungi Recognition: A Practical Use Case
Šulc, M., Picek, L., Matas, J., Jeppesen, T. S., & Heilmann‐Clausen, J. (2020) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv45572.2020.9093624 · Cited by 30
“The topic of adjusting CNN predictions to new priors is discussed in [7, 25, 27]: in the case when the new class priors pe(k) are known, the new posterior pe(k|x) can be computed as:”
BackgroundA unified framework for dataset shift diagnostics
Polo, F. M., Izbicki, R., Lacerda, E. G., Jimenez, J. P. I., & Vicente, R. (2023) · Information Sciences · doi:10.1016/j.ins.2023.119612 · Cited by 31
XLSearch: a Probabilistic Database Search Algorithm for Identifying Cross-Linked Peptides
Ji, C., Li, S., Reilly, J. P., Predrag Radivojac, A., & Tang, H. (2016) · Journal of Proteome Research · doi:10.1021/acs.jproteome.6b00004 · Cited by 20
Systematic Labeling Bias: De-biasing Where Everyone is Wrong
Cabrera-Vives, G., Miller, C. J., & Schneider, J. (2014) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr.2014.756 · Cited by 18
“For example, [6] addressed the problem of class balance by using an EM algorithm which updates the test class-prior and class-posterior probabilities from some initial estimates.”
MethodDistribution-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 107
“In light of the theoretical findings in our paper, we recommend some form of binning as the last step of calibrated prediction due to the robust distribution-free guarantees provided by Theorem 4.”
BackgroundImproving CNN Classifiers by Estimating Test-Time Priors
Šulc, M., & Matas, J. (2019) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw.2019.00402 · Cited by 20
“Methods [3, 14] for adjusting classifier outputs to new and unknown a-priori probabilities on the test set have been published years ago, yet the problem of changed class priors is commonly not addressed in computer vision tasks where the situation arises.”
BackgroundMethodDataset shift assessment measures in monitoring predictive models
Becker, A., & Becker, J. (2021) · Procedia Computer Science · doi:10.1016/j.procs.2021.09.112 · Cited by 22
Classifying documents with link-based bibliometric measures
Couto, T., Ziviani, N., Calado, P. P., de Cristo, M. A. P., Gonçalves, M. A., de Moura, E. S., & Brandão, W. C. (2009) · Information Retrieval · doi:10.1007/s10791-009-9119-7 · Cited by 17
“For instance, in Saerens et al. (2002), a simple iterative procedure was presented to adjust the outputs of a classifier with respect to new a priori probabilities estimated from prediction on actual data.”
MethodConfidence Intervals for Class Prevalences under Prior Probability Shift
Tasche, D. (2019) · Machine Learning and Knowledge Extraction · doi:10.3390/make1030047 · Cited by 21
“[18] made the EM-algorithm version of the approach—as a possibility to obtain the maximum likelihood estimate [32]—popular in the machine learning community.”
BackgroundMethodInfluentialMulti-mission Terrain Classifier for Safe Rover Navigation and Automated Science
Atha, D., Swan, R. M., Didier, A., Hasnain, Z., & Ono, M. (2022) · IEEE Aerospace Conference (AERO) · doi:10.1109/aero53065.2022.9843615 · Cited by 23
Dihge, L., Bendahl, P., Skarping, I. D., Hjärtström, M., Ohlsson, M., & Rydén, L. (2023) · Frontiers in Oncology · doi:10.3389/fonc.2023.1102254 · Cited by 15
“However, procedures to adjust outputs of a classifier to new a priori probability has been proposed to mitigate these errors and might be applied in the future (56).”
BackgroundMonitoring and backtesting churn models
Lima, E. D. O., Mues, C., & Baesens, B. (2010) · Expert Systems with Applications · doi:10.1016/j.eswa.2010.07.091 · Cited by 23
“Saerens, Latinne, and Decaestecker (2002) describe an adjustment procedure for when the population prior churn probabilities are known.”
BackgroundQCRI at SemEval-2016 Task 4: Probabilistic Methods for Binary and Ordinal Quantification
Martino, G. D. S., Gao, W. G., & Sebastiani, F. (2016) · International Workshop on Semantic Evaluation (SemEval) · doi:10.18653/v1/s16-1006 · Cited by 17
“, 2010), Expectation Maximization for Quantification (EMQ) (Saerens et al., 2002), SVMs optimized for KLD (SVM(KLD)) (Esuli and Sebastiani, 2015), SVMs optimized for NKLD”
MethodInfluentialPereira, L. A. A., Frenzel, M., Buchmann, M., Kern, M., Tolosana‐Delgado, R., van den Boogaart, K. G., & Gutzmer, J. (2022) · International Journal of Mining Science and Technology · doi:10.1016/j.ijmst.2022.01.008 · Cited by 14
Sánchez-González, L., González, V., Alegre, E., & Aláiz-Rodríguez, R. (2008) · Lecture notes in computer science · doi:10.1007/978-3-540-69812-8_82 · Cited by 14
Calabrese, R., Osmetti, S. A., & Zanin, L. (2019) · Journal of the Royal Statistical Society Series A (Statistics in Society) · doi:10.1111/rssa.12523 · Cited by 14
“In order to consider well-calibrated probabilities of default compliant to Basel III (Basel Committee on Banking Supervision, 2015), we apply the adjust of the posterior default probabilities proposed in Saerens et al. (2002) and Baesens et al. (2016).”
MethodMixture Proportion Estimation and PU Learning: A Modern ApproachPreprint
Garg, S., Wu, Y., Smola, A. J., Balakrishnan, S., & Lipton, Z. C. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.00980 · Cited by 17
“To begin making this problem tractable, we might make the label shift assumption [37, 41, 29], i.e., that while the class balance p p y q can change, the class conditional distributions p p x | y q do not.”
BackgroundAutomatic Dataset Shift Identification to Support Safe Deployment of Medical Imaging AI
Roschewitz, M., Mehta, R., Jones, C., & Glocker, B. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-04981-0_7 · Cited by 27
“For this, we can leverage the prevalence shift adaptation literature, where various methods have been proposed to estimate the density ratio: with w ∈ R C [1,32,39].”
Binary quantification and dataset shift: an experimental investigation
González, P., Moreo, A., & Sebastiani, F. (2024) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-024-01014-1 · Cited by 25
“A further, very popular aggregative quantification method is the one proposed by Saerens et al. (2002) and often called SLD, from the names of its proposers.”
InfluentialImportance Weight Estimation and Generalization in Domain Adaptation Under Label Shift
Azizzadenesheli, K. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2021.3086060 · Cited by 18
“…et al., 2012, Kull and Flach, 2014, Hofer, 2015, Tasche, 2017], a work by Lipton et al. [2018] propose black box correction method which is applicable to a wide range of label shift problems and drawn from the classical grouping problemBuck et al. [1966], Forman [2008], Saerens et al. [2002].”
Estimating Class Proportions in Boar Semen Analysis Using the Hellinger Distance
González‐Castro, V., Aláiz-Rodríguez, R., Fernández-Robles, L., Guzmán-Martínez, R., & Alegre, E. (2010) · Lecture notes in computer science · doi:10.1007/978-3-642-13022-9_29 · Cited by 13
The Hitchhiker’s Guide to Prior-Shift Adaptation
Sipka, T., Šulc, M., & Matas, J. (2022) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv51458.2022.00209 · Cited by 15
“Table 2 compares accuracy after adaptation with new prior estimate based on confusion matrix (CM) inversion [18] and our proposed method from Section 3.”
BackgroundMethodInfluentialMinimax classifiers based on neural networks
Aláiz-Rodríguez, R., Guerrero-Curieses, A., & Cid‐Sueiro, J. (2004) · Pattern Recognition · doi:10.1016/j.patcog.2004.05.007 · Cited by 13
A Near-Optimal Algorithm for Debiasing Trained Machine Learning ModelsPreprint
Alabdulmohsin, I., & Lučić, M. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.12887 · Cited by 29
“(2016) (2) the shift inference method, first introduced in (Saerens et al., 2002) and used more recently in (Wang et al.”
MethodDeployment of Image Analysis Algorithms Under Prevalence Shifts
Godau, P., Kalinowski, P., Christodoulou, E., Reinke, A., Tizabi, M. D., Ferrer, L., Jäger, P. F., & Maier-Hein, L. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-43898-1_38 · Cited by 21
Navigating prevalence shifts in image analysis algorithm deployment
Godau, P., Kalinowski, P., Christodoulou, E., Reinke, A., Tizabi, M. D., Ferrer, L., Jäger, P. F., & Maier-Hein, L. (2025) · Medical Image Analysis · doi:10.1016/j.media.2025.103504 · Cited by 19
Diagnosing Supercell Environments: A Machine Learning Approach
Shield, S. A., & Houston, A. L. (2022) · Weather and Forecasting · doi:10.1175/waf-d-21-0098.1 · Cited by 19
A Concise Overview of LeQua@CLEF 2022: Learning to Quantify
Esuli, A., Moreo, A., Sebastiani, F., & Sperduti, G. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-13643-6_23 · Cited by 14
Novaes, M. T., de Carvalho, O. L. F., Ferreira, P. H. G., Tiraboschi, T. L. N., Silva, C. S., Contreras, J. C. Z., Gomes, C. M., Miranda, E., Júnior, O. A. D. C., & de Bessa, J. (2021) · Informatics in Medicine Unlocked · doi:10.1016/j.imu.2021.100538 · Cited by 17
Local sparsity control for naive Bayes with extreme misclassification costs
Kołcz, A. (2005) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/1081870.1081888 · Cited by 12
Exact Fit of Simple Finite Mixture Models
Tasche, D. (2014) · Journal of risk and financial management · doi:10.3390/jrfm7040150 · Cited by 20
“In [11], the iteration was derived differently to [10], namely by studying the gradient of the likelihood function.”
BackgroundTransductive Methods for the Distributed Ensemble Classification Problem
Miller, D. J., & Pal, S. (2007) · Neural Computation · doi:10.1162/neco.2007.19.3.856 · Cited by 12
“The IT method can be applied for the case of a single classifier, as considered in Saerens et al. (2002), and in this case, it will not require the statistical assumptions made there.”
BackgroundMethodInfluentialA Comparative Evaluation of Quantification MethodsPreprint
Schumacher, T., Strohmaier, M., & Lemmerich, F. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.03223 · Cited by 43
“…group of algorithms which show a significantly better average performance than the rest, with the Generalized Probabilistic Adjusted Count [3, 14], readme [21], energy distance minimization [23], the EM algorithm for quantification [31], and Friedman’s method [17] leading in the averaged rankings.”
BackgroundMethodHandling Concept Drift via Ensemble and Class Distribution Estimation Technique
Limsetto, N., & Waiyamai, K. (2011) · Lecture notes in computer science · doi:10.1007/978-3-642-25856-5_2 · Cited by 13
“This problem can be solved in various ways such as robust classifier [2, 10], data sampling [5, 15], semi-supervised learning [4] and cost-based learning [1, 9 ,14, 19].”
BackgroundImproving Classification under Changes in Class and Within-Class Distributions
Aláiz-Rodríguez, R., Guerrero-Curieses, A., & Cid‐Sueiro, J. (2009) · Lecture notes in computer science · doi:10.1007/978-3-642-02478-8_16 · Cited by 15
Adapting a classification rule to local and global shift when only unlabelled data are available
Hofer, V. (2014) · European Journal of Operational Research · doi:10.1016/j.ejor.2014.11.022 · Cited by 19
Unified Framework for QuantificationPreprint
Firat, A. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1606.00868 · Cited by 30
“For post-quantification classification we use the post-quantification classification rule by (Saerens et al., 2002):”
MethodDistribution-free uncertainty quantification for classification under label shiftPreprint
Podkopaev, A., & Ramdas, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.03323 · Cited by 138
“A well-known alternative approach to directly estimate the importance weights which performs well in practice is MLLS [22] and its recent variations that combine it with preceding calibration on the source domain [1].”
BackgroundMethodInfluentialA Primer on Domain AdaptationPreprint
Lemberger, P., & Panico, I. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2001.09994 · Cited by 18
“Table 1: Efficiency of the EM method on three real medical data sets used in[1].”
BackgroundMethodInfluentialCoping with Label Shift via Distributionally Robust OptimisationPreprint
Zhang, J., Menon, A., Veit, A., Bhojanapalli, S., Kumar, S., & Sra, S. (2020) · International Conference on Learning Representations · doi:10.48550/arxiv.2010.12230 · Cited by 86
“…to learning with a modified label distribution. the special case of label shift , one posits that p te ( x | y ) = p tr ( x | y ) , but the label distribution p te ( y ) (cid:54) = p tr ( y ) (Saerens et al., 2002); i.e., the test distribution satisfies p te ( x, y ) = p te ( y ) p tr ( x | y ) .”
BackgroundGain-Some-Lose-Some: Reliable Quantification Under General Dataset Shift
Denham, B., Lai, E. M.-K., Sinha, R., & Naeem, M. A. (2021) · Proceedings · doi:10.1109/icdm51629.2021.00121 · Cited by 9
Gan, R. W., Sun, D. K., Tatro, A. R., Cohen‐Mekelburg, S., Wiitala, W. L., Zhu, J., & Waljee, A. K. (2021) · PLoS ONE · doi:10.1371/journal.pone.0257520 · Cited by 11
“Because flare is a rare event, the non-events were randomly under-sampled to match the distribution of events to avoid class imbalance; model predicted probabilities were corrected to account for the rare event proportion using Bayes’ rule [10].”
MethodPlant Recognition by Inception Networks with Test-time Class Prior Estimation.
Šulc, M., Picek, L., & Matas, J. (2018) · CLEF (Working Notes) · Cited by 9
Kim, J., Oh, J., & Heo, T. (2021) · International Journal of Interactive Multimedia and Artificial Intelligence · doi:10.9781/ijimai.2021.08.009 · Cited by 9
Tasche, D. (2021) · Statistics · doi:10.1080/02331888.2021.2016767 · Cited by 12
“2 of Schervish (1989)? Unfortunately, Schervish (1989) did not provide a detailed proof of the theorem but only suggested the building stones needed for a proof. Therefore it seems possible that in some of the steps of the proof an assumption would have to be made that forecaster A had no more information at her dispo…”
BackgroundInfluentialLabel Shift Estimation for Class-Imbalance Problem: A Bayesian Approach
Ye, C., Tsuchida, R., Petersson, L., & Barnes, N. M. (2024) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv57701.2024.00111 · Cited by 9
“We follow previous methods [1, 2, 22] to evaluate label shift estimation performance with ( w − ˆ w ) 2 /K , where w i = P ( Y t = i ) / P ( Y s = i ) , i = 1 , 2 ...”
BackgroundMethodInfluentialCalibration methods in imbalanced binary classification
Guilbert, T., Caelen, O., Chirita, A., & Saerens, M. (2024) · Annals of Mathematics and Artificial Intelligence · doi:10.1007/s10472-024-09952-8 · Cited by 16
Self-citationFukushima, T., Kurokawa, R., Hagiwara, A., Sonoda, Y., Asari, Y., Kurokawa, M., Kanzawa, J., Gonoi, W., & Abe, O. (2025) · Japanese Journal of Radiology · doi:10.1007/s11604-025-01743-3 · Cited by 9
“Note that in the field of machine learning, it is known that the performance of a classifier improves when the correct prior probabilities are input [18].”
Failure Modes of Domain Generalization Algorithms
Galstyan, T., Harutyunyan, H., Khachatrian, H., Steeg, G. V., & Galstyan, A. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.01849 · Cited by 15
“The out-of-distribution (OOD) and robustness literature focus on the case when there is a distribution shift at test time [30, 39], including but not limited to label shifts [22, 34], co-variate shifts [12, 38], conditional shifts [44], visual distortions [7, 16], stylistic and other changes [15].”
BackgroundQuantifying the proportion of damaged sperm cells based on image analysis and neural networks
Aláiz-Rodríguez, R., Alegre, E., González‐Castro, V., & Sánchez-González, L. (2008) · Cited by 8
Uncertainty-aware estimation of population abundance using machine learning
Boom, B., Beauxis-Aussalet, E., Hardman, L., & Fisher, R. B. (2015) · Multimedia Systems · doi:10.1007/s00530-015-0479-0 · Cited by 8
“While in [23], authors estimate the a priori distribution of a new dataset based on the features.”
BackgroundAccuracy analysis of semi-supervised classification when the class balance changes
Yamazaki, K. (2015) · Neurocomputing · doi:10.1016/j.neucom.2014.10.080 · Cited by 8
Constrained Least-Squares Density-Difference Estimation
Nguyen, T. D., du Plessis, M. C., Kanamori, T., & Sugiyama, M. (2014) · IEICE Transactions on Information and Systems · doi:10.1587/transinf.e97.d.1822 · Cited by 8
“We then learn the test class balance by matching a mixture of class-wise training input densities, πptrain(x|y = +1) + (1− π)ptrain(x|y = −1), with the test input density ptest(x) [17, 6].”
MethodPredicting Airplane Go-Arounds Using Machine Learning and Open-Source Data
Figuet, B., Monstein, R., Waltert, M., & Barry, S. I. (2020) · doi:10.3390/proceedings2020059006 · Cited by 8
“, as presented in [19]) is not applied.”
BackgroundA Fixed-Point Algorithm to Minimax Learning With Neural Networks
Guerrero-Curieses, A., Aláiz-Rodríguez, R., & Cid‐Sueiro, J. (2004) · IEEE Transactions on Systems Man and Cybernetics Part C (Applications and Reviews) · doi:10.1109/tsmcc.2004.833284 · Cited by 10
“So, the fixed point attained by the algorithm is the minimax solution searched.”
BackgroundMethodResultInfluentialMiller, D. J., Zhang, Y., & Kesidis, G. (2008) · EURASIP Journal on Advances in Signal Processing · doi:10.1155/2008/674974 · Cited by 9
“This work built on [18], which applied transductive learning to adapt class priors while making decisions in the case of a single classifier.”
BackgroundMethodInfluentialDid the Model Change? Efficiently Assessing Machine Learning API ShiftsPreprint
Chen, L., Cai, T., Zaharia, M., & Zou, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.14203 · Cited by 19
Castro-Martín, L., Rueda, M. D. M., Sánchez-Cantalejo, C., Ferri‐García, R., & Cabrera‐León, A. (2024) · BMC Medical Research Methodology · doi:10.1186/s12874-024-02171-z · Cited by 8
Marsono, M. N. (2007) · Cited by 7
Xu, J., & Huang, S. (2023) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3583780.3615173 · Cited by 13
“For 𝑛 new test observations x 1 , ... x 𝑛 , the likelihood can be formulated by To solve this problem, a simple EM algorithm can be derived with the following steps [32]: M-Step: where 𝜏 is the iteration step index, Eq.”
BackgroundMethodBourget, M., Boudreault, M., Carozza, D. A., Boudreault, J., & Raymond, S. (2024) · Astin Bulletin · doi:10.1017/asb.2024.19 · Cited by 10
Does quantification without adjustments work?Preprint
Tasche, D. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1602.08780 · Cited by 21
A Flexible and Adaptive Framework for Abstention Under Class ImbalancePreprint
Shrikumar, A., Alexandari, A. M., & Kundaje, A. B. (2018) · arXiv (Cornell University) · Cited by 7
Tracking the risk of a deployed model and detecting harmful distribution shiftsPreprint
Podkopaev, A., & Ramdas, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.06177 · Cited by 47
“Commonly studied settings include covariate shift (Shimodaira, 2000) and label shift (Saerens et al., 2002).”
BackgroundChen, L., Zheng, Z., & Wu, D. (2024) · Bioengineering · doi:10.3390/bioengineering11111146 · Cited by 6
“They utilized the EM algorithm [34] to constrain the proportion of each class in segmentation predictions to match its in the unlabelled pixels.”
From Thoughts to Actions: A Longitudinal Examination of NSSI Risk Factors in Adolescence
Zhou, J., Hu, S., Wu, H., Wang, S., & Gong, J. (2025) · Psychology Research and Behavior Management · doi:10.2147/prbm.s545595 · Cited by 10
“55 These approaches can complement Random Forests by modeling within-person dynamics and by explicitly accommodating ambiguity in the thoughts stage.”
Zhang, D., Svensson, M., Edén, P., & Dihge, L. (2024) · Scientific Reports · doi:10.1038/s41598-024-78040-y · Cited by 9
Novel Approach for Predicting COVID-19 Symptoms using ARM based APRIORI Algorithm
Babu, S. A., Raj, R. J. S., Varalatchoumy, M., Gopila, M., & Justin, B. V. F. (2022) · International Conference on Computing Methodologies and Communication (ICCMC) · doi:10.1109/iccmc53470.2022.9753987 · Cited by 7
Xia, Y., Liao, Z., Xu, J., & Li, Y. (2022) · Technological and Economic Development of Economy · doi:10.3846/tede.2022.17045 · Cited by 8
Moreo, A., Manuel, F. J. S., & Sebastiani, F. (2023) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/3606264 · Cited by 15
“Afurther, very popular (aggregative) quantification method is the one proposed in [50], which is often called SLD, from the names of its proposers, and which was called EMQ in [22].”
MethodLabel Shift Quantification with Robustness Guarantees via Distribution Feature Matching
Dussap, B., Blanchard, G., & Chérief-Abdellatif, B.-E. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-43424-2_5 · Cited by 14
Hung, C.-M., & Huang, Y. (2006) · Expert Systems with Applications · doi:10.1016/j.eswa.2006.08.018 · Cited by 7
Revisiting Distributionally Robust Supervised Learning in ClassificationPreprint
Hu, W., Niu, G., Sato, I., & Sugiyama, M. (2016) · arXiv (Cornell University) · Cited by 8
Shrikumar, A., & Kundaje, A. B. (2019) · Cited by 5
Threshold prediction for detecting rare positive samples using a meta-learner
Zefrehi, H. G., Sheikhi, G., & Altınçay, H. (2022) · Pattern Analysis and Applications · doi:10.1007/s10044-022-01103-1 · Cited by 5
“suggest a transformation to adjust the outputs of classifiers to improve classification performance in the case of having different a priori probabilities in the training and test data [39].”
BackgroundClassification from positive and unlabeled data based on likelihood invariance for measurement
Yoshida, T., Washio, T., Ohshiro, T., & Taniguchi, M. (2021) · Intelligent Data Analysis · doi:10.3233/ida-194980 · Cited by 5
“Saerens et al. proposed a covariate shift adaptation for the difference in the class prior in the semi-supervised setting [22].”
Wang, X. Y., Biswas, M. K., & Raghupathy, S. (2007) · Sketch Based Interfaces and Modeling · doi:10.1145/1384429.1384458 · Cited by 5
“Following the example of Saerens et al [ SLD02 ], we describe how to adjust the outputs of the neural network to a new file to produce more accurate estimates of the posterior probabilities, p(di|xi).”
InfluentialMinimax optimal approaches to the label shift problemPreprint
Maity, S., Sun, Y., & Banerjee, M. (2020) · arXiv (Cornell University) · Cited by 5
SALτ: efficiently stopping TAR by improving priors estimates
Molinari, A., & Esuli, A. (2023) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-023-00961-5 · Cited by 10
“SAL 𝜏 is a variant of the well-known SLD algo-rithm (Saerens et al. 2002): our algorithm was designed to enable the use of SLD in AL and TAR processes, solving the issues observed in Esuli et al. (2022) and Moli-nari et al. (2023).”
Information Gain Sampling for Active Learning in Medical Image Classification
Mehta, R., Shui, C., Nichyporuk, B., & Arbel, T. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-16749-2_13 · Cited by 10
Hjärtström, M., Dihge, L., Bendahl, P., Skarping, I. D., Ellbrant, J., Ohlsson, M., & Rydén, L. (2023) · JMIR Cancer · doi:10.2196/46474 · Cited by 3
“Model predictions were recalibrated to the prevalence in the external validation cohort to account for the different N status distributions of cohorts I and II [32].”
MethodA general framework for branch length estimation in Ancestral Recombination Graphs
Deng, Y. K., DeWitt, W. S., Song, Y. S., & Nielsen, R. (2025) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2504461122 · Cited by 7
STRATEGIES FOR QUALITY-AWARE VIDEO CONTENT ANALYTICS
Reibman, A. R. (2018) · IEEE Southwest Symposium on Image Analysis and Interpretation (SSIAI) · doi:10.1109/ssiai.2018.8470354 · Cited by 4
“Other approaches include domain adaptation and transfer learning [42–46], using data augmentation during training to include additional synthetically created distortions, or simply training the system on low-quality data [10].”
BackgroundAspect-Based Sentiment Quantification
Matsiiako, V., Frăsincar, F., & Boekestijn, D. (2022) · IEEE Transactions on Affective Computing · doi:10.1109/taffc.2022.3218504 · Cited by 4
“One of the initial works in the field of quantification was published by [27].”
BackgroundMethodMC-SQ and MC-MQ: Ensembles for Multi-Class Quantification
Donyavi, Z., Serapião, A. B. D. S., & Batista, G. (2024) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2024.3372011 · Cited by 6
Quantification using permutation-invariant networks based on histograms
Pérez-Mon, O., Moreo, A., del Coz, J. J., & González, P. (2024) · Neural Computing and Applications · doi:10.1007/s00521-024-10721-1 · Cited by 11
“…(PCC): and Probabilistic Adjusted Classify and Count (PACC) (also known as BBSE-soft in [24]): The Expectation Maximization for Quantification (EMQ) [28] method applies the EM algorithm to adjust the posterior probabilities generated by a soft classifier s to the potential shift in the label…”
Adapting to Label Shift with Bias-Corrected Calibration
Shrikumar, A., Alexandari, A. M., & Kundaje, A. B. (2019) · Cited by 4
Shift Happens: Adjusting Classifiers
Heiser, T. J. T., Allikivi, M.-L., & Kull, M. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-46147-8_4 · Cited by 6
Automatic Assignment of Non-Leaf MeSH Terms to Biomedical Articles.
Kavuluru, R., & Rios, A. (2015) · PubMed Central · Cited by 4
NRU-HSE at SemEval-2017 Task 4: Tweet Quantification Using Deep Learning Architecture
Karpov, N. (2017) · International Workshop on Semantic Evaluation (SemEval) · doi:10.18653/v1/s17-2113 · Cited by 4
“A simple procedure to adjust the outputs of a classifier to a new a priori probability is described in the study by (Saerens et al., 2002).”
MethodKontos, K., Godard, P., André, B., van Helden, J., & Bontempi, G. (2008) · BMC Proceedings · doi:10.1186/1753-6561-2-s4-s5 · Cited by 4
“Therefore, we adjust the posterior probabilities returned by the classifiers with respect to new prior probabilities by Bayes' theorem [13].”
MethodA Posteriori Corrections to Classification Methods
Duch, W., & Itert, Ł. (2003) · Physica-Verlag eBooks · doi:10.1007/978-3-7908-1902-1_61 · Cited by 4
The Law of Total OddsPreprint
Tasche, D. (2013) · arXiv (Cornell University) · doi:10.48550/arxiv.1312.0365 · Cited by 4
“There is no need to deploy the expectation-maximisation or other more ad- vance iterative algorithms as discussed by Saerens et al. (2002), Xue and Weiss (2009) or Du Plessis and Sugiyama (2014).”
BackgroundMethodInfluentialQuantificationLib: A Python library for quantification and prevalence estimation
Castaño, A., Alonso, J., González, P., Pérez-Núñez, P., & del Coz, J. J. (2024) · SoftwareX · doi:10.1016/j.softx.2024.101728 · Cited by 7
CV.eDNA: A hybrid approach to invertebrate biomonitoring using computer vision and DNA metabarcodingPreprint
Blair, J. D., Weiser, M. D., Siler, C. D., Kaspari, M. E., Smith, S. N., McLaughlin, J. F., & Marshall, K. E. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.09.02.610558 · Cited by 5
“If a priori probabilities for classes in a classification model are known, the outputs of the model can be adjusted according to those probabilities to improve classification accuracy (Saerens, Latinne and Decaestecker, 2002).”
Cheng, J., Tian, J., Spoto, F., Azhir, A., Mork, D., & Estiri, H. (2025) · Scientific Reports · doi:10.1038/s41598-025-32656-w · Cited by 9
Eusebi, R., Su, H., Wu, L., Rong, P., Balaguru, K., Leung, L.-Y. R., Choi, Y., Chan, P. W., Gan, J., DeMaria, M., et al. (2025) · Environmental Research Letters · doi:10.1088/1748-9326/adac7f · Cited by 6
“Instead of a calibrator, the uncalibrated model still has its probabilities transformed to be representative of a distribution with the true average RI rate, not a 50% rate, as described in Saerens et al (2002).”
Assessing generalizability of a dengue classifier across multiple datasets
Lu, B., Li, Y., & Evans, C. (2025) · PLoS ONE · doi:10.1371/journal.pone.0323886 · Cited by 3
“Conversely, label shift [37–39] assumes that the distribution of the response (i.e., dengue prevalence) changes between datasets, but the distributions of age, white blood cell count, and platelet count remain the same conditional on dengue status.”
InfluentialInterpretable domain adaptation using unsupervised feature selection on pre-trained source models
Zhang, L., Germain, P., Kessaci, Y., & Christophe Biernacki, A. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.09.031 · Cited by 3
On the Influence of Data Imbalance on Supervised Gaussian Mixture Models
Scrucca, L. (2023) · Algorithms · doi:10.3390/a16120563 · Cited by 3
Metadata
- Authors
- Saerens, M., Latinne, P., & Decaestecker, C.
- Affiliations
- Université Libre de Bruxelles (BE)
- Venue
- Neural Computation
- Publisher
- MIT Press
- Published
- 2002-01-01
- Volume
- 14
- Issue
- 1
- Pages
- 21-41
- ISSN
- 0899-7667 (ISSN-L)
- 1530-888X
- OpenAlex
- W2141253686
- Cited by
- 361
- References
- 26
- Citations by year
- 2026: 57 · 2025: 28 · 2024: 26 · 2023: 24 · 2022: 23 · 2021: 32 · 2020: 27 · 2019: 19 · 2018: 15 · 2017: 9 · 2016: 17 · 2015: 11 · 2014: 9 · 2013: 4 · 2012: 6
- FWCI
- 4.14
- Type
- journal article
- Language
- en
- Open access
- green
- dipot.ulb.ac.be
- Licence
- other-oa
- Locations
- Neural Computation (publishedVersion)
- PubMed (publishedVersion)
- Digital Access to Libraries (Université catholique de Louvain (UCL), l'Université de Namur (UNamur) and the Université Saint-Louis (USL-B)) (submittedVersion)
- Dépôt institutionnel de l'Université libre de Bruxelles (Université Libre de Bruxelles) (submittedVersion, other-oa)
- 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 › Data Stream Mining Techniques
- Keywords
- posterior probability, expectation-maximization algorithm, likelihood maximization, multilayer neural networks, case-control study, classification accuracy
- MeSH terms
- Classification, Probability, Probability Theory
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.