Pith. sign in

REVIEW 4 major objections 4 minor 97 references

Imbalanced Regression Pipeline Recommendation

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A meta-learner beats 42 fixed imbalanced-regression pipelines

desk verdict A useful first meta-learning recommender for imbalanced regression pipelines, with real code/data and an extensive benchmark, but the zero-shot evaluation leaks target-dependent meta-features and the abstract overstates significance. read the letter →

arxiv 2507.11901 v1 pith:DUVHIK6V submitted 2025-07-16 cs.LG

Juscimara G. Avelino , George D. C. Cavalcanti , Rafael M. O. Cruz This is my paper · ORCID
classification cs.LG
keywords imbalancedregressionmeta-learningpipelinerecommendationresamplingstrategieszero-shotF1-scoreRSERAAutoMLcomparison
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that the best pipeline for an imbalanced regression dataset—defined as the pair of a resampling strategy and a regression model—can be chosen instantly from dataset meta-features, without trying dozens of configurations. It introduces Meta-IR, a meta-learning framework that trains two classifiers on a meta-dataset of 218 imbalanced regression problems: one recommends the resampling strategy, the other the learning model. The central finding is that Meta-IR's recommendations beat all 42 fixed combinations of six resampling strategies (including none) and six regression models, and also beat six general-purpose AutoML frameworks, on metrics that emphasize rare target values. A chained training variant, where one recommendation feeds the other, improves accuracy and reveals that the choice of learning model and resampling strategy are coupled. If the claim holds, practitioners get a fast, zero-shot alternative to exhaustive pipeline search, at roughly 50 times lower online cost than AutoML systems.

What carries the argument

The engine is the meta-dataset M, built from 218 datasets: each dataset Di is represented by a vector of 43 meta-features (simple counts such as number of samples, attributes, rare cases and percentage of rare cases, plus complexity measures), and by a meta-target (li, ri) equal to the pipeline with the best F1-scoreR or SERA score. Two meta-classifiers, λL and λR, are trained on M using Random Forest. In the Independent scheme they are trained separately; in the Chained schemes, one meta-classifier's prediction is appended as an extra meta-feature before training the other, yielding Model First and Strategy First variants. The imbalance itself is defined through a relevance function φ that maps each continuous target to [0,1] using pchip interpolation over Tukey-based control points, with a threshold tR = 0.8 splitting rare from normal examples. This machinery encodes 'what kind of imbalance does this dataset have' and maps it to a pipeline recommendation in one prediction step.

What would settle it

Re-run the leave-one-dataset-out comparison with meta-features extracted only from the training split of each dataset and check whether Meta-IR still beats all 42 fixed pipelines and the AutoML frameworks on F1-scoreR and SERA; if the advantage shrinks or reverses, the reported zero-shot gains depend on test-target leakage.

Watch

Extended reading notes

Core claim

The paper's core discovery is that a pair of meta-classifiers, trained on meta-features of previously seen imbalanced regression problems, can identify the pipeline that a domain-specific metric will prefer on a new problem, and that this identification is better than relying on any fixed pipeline or on general AutoML search. Over 218 datasets, with six regression models and six resampling strategies plus no resampling, the recommended pipelines (using the Chained 'Model First' variant) won more datasets than any of the 42 fixed configurations and were statistically significantly better than 40 of them on F1-scoreR and better than all of them on SERA. Against Auto-Sklearn, H2O, TPOT, FLAML, LightAutoML and NaiveAutoML, Meta-IR won or tied on a majority of datasets, was significantly better on F1-scoreR overall, and was significantly better on SERA in the high-rare-cases subgroup; the recommendation phase ran about 50 times faster. The authors read this as evidence that there is no one-size-fits-all pipeline for imbalanced regression, and that the dataset's imbalance structure—particularly the percentage of rare target cases—carries enough information to make the choice in a zero-shot way.

Load-bearing premise

The load-bearing premise is that the meta-features used to recommend a pipeline can be computed at recommendation time, but the two most important ones—the number and percentage of rare target values—are computed in this study from the whole dataset, including the test split, so in real use they might not be available.

Editorial extensions

If this is right

  • A new imbalanced regression dataset can get a pipeline recommendation from roughly 43 meta-features and two pre-trained classifiers, replacing a search over 42 configurations.
  • Because chained training improves recommendation accuracy, the choice of resampling strategy and learning model should not be treated as independent in pipeline search.
  • Datasets with a high percentage of rare target values (above roughly 15%) are where Meta-IR's advantage over AutoML is largest.
  • Adding Meta-IR's resampling recommendation as a preprocessing step raised F1-scoreR for every AutoML framework tested, so AutoML search spaces are missing a useful component.
  • The zero-shot recommendation phase costs about 50 times less time than AutoML runs, making it practical for repeated or large-scale use.

Reading between the lines

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

  • The zero-shot claim depends on meta-features that include test-target statistics such as n.rare and p.rare computed on the whole dataset; in a true deployment on a new target, these must be estimated from training data or proxy statistics, and the gap between those two settings is not measured in the paper.
  • The chaining recipe—use one recommendation as a meta-feature for the other—is transferable to other pipeline choices where components interact, such as scaling plus model or imputation plus model.
  • A natural extension is to recommend a ranked list of pipelines rather than a single winner; ranking would hedge against meta-classifier errors and let users trade performance for runtime, which the paper lists as future work but does not test.
  • The strong performance on high-rare-case datasets suggests the method should be stress-tested on even more extreme imbalance distributions, such as p.rare below 2%, and on time series, where resampling interacts with temporal structure.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes Meta-IR, a meta-learning framework that recommends a pipeline (one of six learning algorithms and one of six resampling strategies or no resampling) for imbalanced regression tasks. Two recommendation formulations are introduced: Independent, which trains separate meta-classifiers for the learning model and the resampling strategy, and Chained, which feeds the output of one meta-classifier into the other in two possible orders (Model First and Strategy First). The framework is evaluated with leave-one-dataset-out on 218 datasets, comparing Meta-IR with 42 fixed pipelines, Random and Majority baselines, and six AutoML frameworks, using F1-scoreR and SERA. The paper claims that the Chained formulation performs best, that Meta-IR outperforms all fixed pipelines at the base level, that it beats the AutoML frameworks on F1-scoreR and on high-rare-case SERA scenarios, and that it is roughly 50 times faster than AutoML frameworks.

Significance. If the evaluation were sound, the paper would make a useful contribution: it is, to my knowledge, the first meta-learning study for imbalanced regression pipeline recommendation, it uses a large and diverse corpus of 218 datasets, and it makes code and data available. The comparison with AutoML frameworks and the analysis of meta-feature importance are also valuable. However, the central empirical claims are currently not established because the meta-features used for zero-shot recommendation are computed on the full datasets, including the test targets, and because several headline statements are stronger than the reported statistical results. The proposed approach is not circular in the sense that the meta-targets are learned from evaluated pipeline performance, but the test-target leakage in the meta-features directly undermines the zero-shot claim and the fairness of the base-level comparisons.

major comments (4)
  1. [§3.2.1, Algorithm 1 line 16; §3.2.3, Algorithm 3 line 3] The meta-features are extracted from the entire dataset Di, including the test split, as shown by fDi ← ExtractMetaFeatures(Di) in Algorithm 1 line 16, and the same is done for a new dataset G in Algorithm 3 line 3. Many of the 43 meta-features are target-dependent, not only n.rare and p.rare but also the C, L, and S complexity measures from Appendix B, all of which use the target values. In the leave-one-dataset-out protocol, the meta-classifier at recommendation time therefore receives statistics computed from the very test targets whose prediction quality is then scored. The 42 fixed pipelines and the AutoML frameworks receive no equivalent access to test targets. This makes the comparisons in Tables 3–6 and Figure 8 not apples-to-apples and invalidates the 'zero-shot' framing. The experiments should be rerun with meta-features computed only from the training split, or the paper should clearly specify what information is assumed available at recommendation time.
  2. [Abstract; §5.2.1, Tables 3 and 4; §5.2.2, Figure 8] The abstract and conclusion state that 'Meta-IR outperformed all of them' with respect to the 42 fixed configurations, and §5.2.1 says Meta-IR 'consistently outperforms all combinations.' This is contradicted by Table 3, where the Wilcoxon test gives p = 0.323 for DT.GN and p = 0.177 for DT.RO under F1-scoreR, i.e., two non-significant comparisons. Similarly, the text in §5.2.2 acknowledges that for SERA, Meta-IR is not significantly different from Majority, H2O, FLAML, and TPOT in the critical-difference diagram. The claims should be weakened to name the specific comparisons that are statistically significant, and the abstract/conclusion should be aligned with those results.
  3. [§3.2.2, Algorithm 2 lines 12–14 and 20–22] The Chained formulations add the prediction of the first meta-classifier as a meta-feature for the second, but those predictions are computed on the same training meta-dataset used to fit the first meta-classifier. For example, in Model First, λL is trained on x and then λL(fDi) is evaluated on each fDi in x. These are in-sample predictions, which are typically overconfident and more accurate than out-of-sample predictions. At recommendation time for a new dataset, however, the chained feature comes from an out-of-sample prediction. This train/test mismatch can inflate the reported advantage of Chained over Independent, which is one of the paper's main claims. The chained meta-features should be generated with out-of-fold or cross-validated predictions during meta-training.
  4. [§5.3, Figure 10] The claim that Meta-IR is 'approximately 50 times faster' than AutoML frameworks is not supported by the reported protocol. The execution-time comparison uses only ten datasets, with no statement of hardware, measurement methodology, whether meta-feature extraction time is included, or how the 50× factor is computed. Since AutoML frameworks were given a 1-hour budget but may finish earlier, the reported times in Figure 10 need to be accompanied by a precise timing protocol and per-system breakdown. This is a load-bearing claim in the paper's list of contributions.
minor comments (4)
  1. [Throughout] The text contains several typos and spacing errors, including 'T able', 'bloxpot', 'pairs of of learning algorithms', and 'return the the recommended learning model'. These should be corrected in a revision.
  2. [§5.2.2, Table 7] The win/tie/loss counts in Table 7 show a substantial number of losses (e.g., 75 for Auto-sklearn and H2O under F1-scoreR), which is difficult to reconcile with the sentence in the same section that Meta-IR 'consistently outperformed' the AutoML frameworks. Please clarify that wins are aggregated across datasets and do not imply superiority on every dataset.
  3. [Appendix A and Appendix B] The appendix tables are numbered inconsistently: Appendix A is labeled 'Table 10' and Appendix B is labeled 'Table 11', but Section 4.2 refers to 'Appendix B, Table 11' while the text also mentions 'Table 9' for the time datasets. Please renumber all tables and cross-references consistently.
  4. [§4.6] The evaluation methodology section says that the meta-level analysis includes comparison with AutoML frameworks, but the actual AutoML comparison is presented in the base-level analysis (§5.2.2). This organizational mismatch should be fixed for clarity.

Circularity Check

0 steps flagged · score 1.0 of 10

No construction-level circularity: the meta-learning derivation is supervised mapping from dataset meta-features to empirically defined best pipelines; the main validity concern is that target-dependent meta-features are extracted from the full dataset including the test split, undermining the 'zero-shot' framing without making the prediction identical to its inputs by construction.

full rationale

The derivation chain is not circular at the equation level. Algorithm 1 defines the meta-target as the pipeline with the best score on the test split ('score ← Evaluate(m, S, metric)', line 11, and the argmax at line 15) while also computing 'fDi ← ExtractMetaFeatures(Di)' (line 16) on the full dataset Di that still contains the test split S created at line 5. Algorithm 3 repeats this for a new dataset: 'fG ← ExtractMetaFeatures(G)' (line 3). Since the meta-feature set includes target-dependent statistics (n.rare, p.rare, C2-C4 correlations with the output, and S1-S4 smoothness/error measures computed from y), the recommendation input is computed from the same target values whose prediction quality is then scored by F1-scoreR/SERA. This is a serious test-target leakage that makes the claimed 'zero-shot' comparisons with AutoML frameworks unfair, but it does not make the recommended (l*, r*) equal to the argmax meta-target by construction: the meta-classifier still learns a mapping and can be wrong. The only self-citations are motivational or related-work references (Avelino et al., 2024, for the no-single-best-pipeline and resampling/model interdependence claims; de Amorim et al., 2024, for scaling meta-learning); the no-single-best claim is independently visible in the paper's own Figure 7 oracle counts, so these citations are not load-bearing. The Chained model's use of its own prediction as a feature is a stacking-style modeling choice, not a derivation from the target; Algorithm 3's omission of the chained augmentation step is a documentation inconsistency rather than circularity. Overall, no central claim reduces to its own inputs by construction; the score reflects the minor self-citation and the target-leak protocol concern, not derivation-level circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the relevance function definition, the availability of target-dependent meta-features, the single-split evaluation of pipelines, and the choice of default hyperparameters. These are all domain assumptions taken from prior work rather than fitted parameters, with the exception of the meta-model and chaining order selected on the same experimental data.

free parameters (2)
  • Meta-classifier algorithm = Random Forest
    Appendix C shows RF achieved the best meta-level accuracy among seven candidates; this model selection is performed on the same 218 datasets, so it is a hand-picked choice that could overfit the meta-level evaluation.
  • Chaining order for Meta-IR = Model First
    Section 5.2.1 states Meta-IR applied in this analysis is Model First, chosen because it is the most effective approach overall, based on the same experimental results.
assumptions (4)
  • domain assumption The relevance function (pchip interpolation over Tukey boxplot control points) correctly identifies rare and normal cases for imbalanced regression.
    Section 2.1.1 establishes this definition from prior work; all meta-features and resampling strategies depend on it.
  • domain assumption Meta-features extracted from the full dataset, including target-dependent statistics such as p.rare and n.rare, are available at recommendation time in zero-shot deployment.
    Algorithm 1 line 16 computes meta-features from the whole dataset D_i, which includes the test split; this is not available when only training data are present.
  • domain assumption Evaluating pipelines on a single train/test split per dataset identifies the best pipeline for that dataset.
    Algorithm 1 uses one split per dataset; no cross-validation is used, so the meta-target may be noisy.
  • domain assumption Default hyperparameters for all six learning models and six resampling strategies are sufficient to assess pipeline quality.
    Sections 4.3 and 4.4 use default hyperparameters; the paper explicitly acknowledges hyperparameter optimization is out of scope.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Imbalanced Regression Pipeline Recommendation." pith.science (2026). https://pith.science/paper/DUVHIK6V

@misc{pith2026250711901,
  author       = {Pith},
  title        = {Pith review of: Imbalanced Regression Pipeline Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DUVHIK6V}},
  note         = {Machine review of arXiv:2507.11901}
}
read the original abstract

Imbalanced problems are prevalent in various real-world scenarios and are extensively explored in classification tasks. However, they also present challenges for regression tasks due to the rarity of certain target values. A common alternative is to employ balancing algorithms in preprocessing to address dataset imbalance. However, due to the variety of resampling methods and learning models, determining the optimal solution requires testing many combinations. Furthermore, the learning model, dataset, and evaluation metric affect the best strategies. This work proposes the Meta-learning for Imbalanced Regression (Meta-IR) framework, which diverges from existing literature by training meta-classifiers to recommend the best pipeline composed of the resampling strategy and learning model per task in a zero-shot fashion. The meta-classifiers are trained using a set of meta-features to learn how to map the meta-features to the classes indicating the best pipeline. We propose two formulations: Independent and Chained. Independent trains the meta-classifiers to separately indicate the best learning algorithm and resampling strategy. Chained involves a sequential procedure where the output of one meta-classifier is used as input for another to model intrinsic relationship factors. The Chained scenario showed superior performance, suggesting a relationship between the learning algorithm and the resampling strategy per task. Compared with AutoML frameworks, Meta-IR obtained better results. Moreover, compared with baselines of six learning algorithms and six resampling algorithms plus no resampling, totaling 42 (6 X 7) configurations, Meta-IR outperformed all of them. The code, data, and further information of the experiments can be found on GitHub: https://github.com/JusciAvelino/Meta-IR.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

97 extracted references · 68 canonical work pages

  1. [1]

    write newline

    " write newline " cite write " FUNCTION editor.postfix editor num.names #1 > "( )" "( )" if FUNCTION editor.trans.postfix editor num.names #1 > "( )" "( )" if FUNCTION trans.postfix translator num.names #1 > "( )" "( )" if FUNCTION authors.editors.reflist.apa5 'field := 'dot := field num.names 'numnames := numnames 'format.num.names := format.num.names na...

  2. [2]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year eprint archive archivePrefix primaryClass adsurl adsnote version label INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.sta...

  3. [3]

    write newline

    " write newline "" before.all 'output.state := FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap duplicate "7" = swap duplicate "8" = swap "9" = or or or or or or or or or FUNCTION n.separate 't := "" #0 'numnames := t empty not t #-1 #1 subs...

  4. [4]

    Pattern Recognition Letters 128:480--487

    Aguiar GJ, Mantovani RG, Mastelini SM, et al (2019) A meta-learning approach for selecting image segmentation algorithm. Pattern Recognition Letters 128:480--487

    2019

  5. [5]

    Information Sciences 584:665--684

    Aguiar GJ, Santana EJ, de Carvalho AC, et al (2022) Using meta-learning for multi-target regression. Information Sciences 584:665--684

    2022

  6. [6]

    Data Mining and Knowledge Discovery 35:2389--2466

    Aminian E, Ribeiro RP, Gama J (2021) Chebyshev approaches for imbalanced data streams regression models. Data Mining and Knowledge Discovery 35:2389--2466

    2021

  7. [7]

    IEEE Transactions on Neural Networks and Learning Systems 36:4805--4819

    de Amorim LB, Cavalcanti GD, Cruz RM (2024) Meta-scaler: A meta-learning framework for the selection of scaling techniques. IEEE Transactions on Neural Networks and Learning Systems 36:4805--4819

    2024

  8. [8]

    Artificial Intelligence Review 57(4):82

    Avelino JG, Cavalcanti GD, Cruz RM (2024) Resampling strategies for imbalanced regression: a survey and empirical analysis. Artificial Intelligence Review 57(4):82

    2024

Show all 97 references
  1. [9]

    Release 97(1):B1

    B \"a ck T, Fogel DB, Michalewicz Z (1997) Handbook of evolutionary computation. Release 97(1):B1

    1997

  2. [10]

    In: 2018 International Joint Conference on Neural Networks, IEEE, pp 1--8

    Barella VH, Garcia LP, de Souto MP, et al (2018) Data complexity measures for imbalanced classification tasks. In: 2018 International Joint Conference on Neural Networks, IEEE, pp 1--8

    2018

  3. [11]

    In: Brazilian Conference on Intelligent Systems, Springer, pp 498--512

    Barella VH, Garcia LP, de Carvalho AC (2020) Simulating complexity measures on imbalanced datasets. In: Brazilian Conference on Intelligent Systems, Springer, pp 498--512

    2020

  4. [12]

    Information Sciences 553:83--109

    Barella VH, Garcia LP, de Souto MC, et al (2021) Assessing the data complexity of imbalanced datasets. Information Sciences 553:83--109

    2021

  5. [13]

    ACM SIGKDD Explorations Newsletter 6(1):20--29

    Batista GE, Prati RC, Monard MC (2004) A study of the behavior of several methods for balancing machine learning training data. ACM SIGKDD Explorations Newsletter 6(1):20--29

    2004

  6. [14]

    Neural Computation 12(8):1889--1900

    Bengio Y (2000) Gradient-based optimization of hyperparameters. Neural Computation 12(8):1889--1900

    2000

  7. [15]

    Journal of Machine Learning Research 13(1):281--305

    Bergstra J, Bengio Y (2012) Random search for hyper-parameter optimization. Journal of Machine Learning Research 13(1):281--305

    2012

  8. [16]

    In: NIPS Workshop on Bayesian Optimization

    Bergstra J, Bardenet R, K \'e gl B, et al (2011) Implementations of algorithms for hyper-parameter optimization. In: NIPS Workshop on Bayesian Optimization

    2011

  9. [17]

    arXiv preprint ArXiv:1604.08079

    Branco P, Ribeiro RP, Torgo L (2016) Ubl: an r package for utility-based learning. arXiv preprint ArXiv:1604.08079

    2016 arXiv

  10. [18]

    In: First International Workshop on Learning with Imbalanced Domains: Theory and Applications, pp 36--50

    Branco P, Torgo L, Ribeiro RP (2017) Smogn: a pre-processing approach for imbalanced regression. In: First International Workshop on Learning with Imbalanced Domains: Theory and Applications, pp 36--50

    2017

  11. [19]

    Neurocomputing 343:76--99

    Branco P, Torgo L, Ribeiro RP (2019) Pre-processing approaches for imbalanced distributions in regression. Neurocomputing 343:76--99

    2019

  12. [20]

    Springer Science & Business Media, Heidelberg

    Brazdil P, Carrier CG, Soares C, et al (2008) Metalearning: Applications to data mining. Springer Science & Business Media, Heidelberg

    2008

  13. [21]

    Springer Nature, Cham

    Brazdil P, van Rijn JN, Soares C, et al (2022) Metalearning: Applications to automated machine learning and data mining. Springer Nature, Cham

    2022

  14. [22]

    In: European Conference on Machine Learning

    Brazdil PB, Soares C (2000) A comparison of ranking methods for classification algorithm selection. In: European Conference on Machine Learning. Springer, pp 63--75

    2000

  15. [23]

    Machine Learning 50:251--277

    Brazdil PB, Soares C, Da Costa JP (2003) Ranking learning algorithms: Using ibl and meta-learning on accuracy and time results. Machine Learning 50:251--277

    2003

  16. [24]

    Machine Learning 24(2):123--140

    Breiman L (1996) Bagging predictors. Machine Learning 24(2):123--140

    1996

  17. [25]

    Machine Learning 45(1):5--32

    Breiman L (2001) Random forests. Machine Learning 45(1):5--32

    2001

  18. [26]

    Routledge

    Breiman L (2017) Classification and regression trees. Routledge

    2017

  19. [27]

    Expert Systems with Applications 193:116387

    Camacho L, Douzas G, Bacao F (2022) Geometric smote for regression. Expert Systems with Applications 193:116387

    2022

  20. [28]

    In: 2012 IEEE 24th International Conference on Tools with Artificial Intelligence, IEEE, pp 1065--1069

    Cavalcanti GD, Ren TI, Vale BA (2012) Data complexity measures and nearest neighbor classifiers: a practical analysis for meta-learning. In: 2012 IEEE 24th International Conference on Tools with Artificial Intelligence, IEEE, pp 1065--1069

    2012

  21. [29]

    Journal of Artificial Intelligence Research 16:321--357

    Chawla NV, Bowyer KW, Hall LO, et al (2002) Smote: synthetic minority over-sampling technique. Journal of Artificial Intelligence Research 16:321--357

    2002

  22. [30]

    In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

    Chen T, Guestrin C (2016) Xgboost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, pp 785--794

    2016

  23. [31]

    Applied Soft Computing 133:109924

    de Amorim LB, Cavalcanti GD, Cruz RM (2023) The choice of scaling technique matters for classification performance. Applied Soft Computing 133:109924

    2023

  24. [32]

    Journal of Machine Learning Research 7(1):1--30

    Dem s ar J (2006) Statistical comparisons of classifiers over multiple data sets. Journal of Machine Learning Research 7(1):1--30

    2006

  25. [33]

    Mathematics of Computation 52(186):471--494

    Dougherty RL, Edelman AS, Hyman JM (1989) Nonnegativity-, monotonicity-, or convexity-preserving cubic and quintic hermite interpolation. Mathematics of Computation 52(186):471--494

    1989

  26. [34]

    Journal of Machine Learning Research 10(15):405--440

    Escalante HJ, Montes M, Sucar LE (2009) Particle swarm model selection. Journal of Machine Learning Research 10(15):405--440

    2009

  27. [35]

    In: Advances in Neural Information Processing Systems, vol 28

    Feurer M, Klein A, Eggensperger K, et al (2015) Efficient and robust automated machine learning. In: Advances in Neural Information Processing Systems, vol 28. Curran Associates, Inc

    2015

  28. [36]

    International Journal of Emerging Technology and Advanced Engineering 2(4):42--47

    Ganganwar V (2012) An overview of classification algorithms for imbalanced datasets. International Journal of Emerging Technology and Advanced Engineering 2(4):42--47

    2012

  29. [37]

    Neurocomputing 176:14--25

    Garcia LP, de Carvalho AC, Lorena AC (2016) Noise detection in the meta-learning level. Neurocomputing 176:14--25

    2016

  30. [38]

    In: 2018 24th International Conference on Pattern Recognition, IEEE, pp 874--879

    Garcia LP, Lorena AC, de Souto MC, et al (2018) Classifier recommendation using data complexity measures. In: 2018 24th International Conference on Pattern Recognition, IEEE, pp 874--879

    2018

  31. [39]

    Information Sciences 445:22--37

    Garc \' a S, Zhang ZL, Altalhi A, et al (2018) Dynamic ensemble selection for multi-class imbalanced datasets. Information Sciences 445:22--37

    2018

  32. [40]

    Cambridge University Press, Cambridge

    Garnett R (2023) Bayesian Optimization. Cambridge University Press, Cambridge

    2023

  33. [41]

    Journal of Big Data 10(1):14

    Garouani M, Ahmad A, Bouneffa M, et al (2023) Autoencoder-knn meta-model based data characterization approach for an automated selection of ai algorithms. Journal of Big Data 10(1):14

    2023

  34. [42]

    Machine Learning 63:3--42

    Geurts P, Ernst D, Wehenkel L (2006) Extremely randomized trees. Machine Learning 63:3--42

    2006

  35. [43]

    Pattern Analysis and Applications 26(1):289--306

    Ghaderi Zefrehi H, Sheikhi G, Alt n c ay H (2023) Threshold prediction for detecting rare positive samples using a meta-learner. Pattern Analysis and Applications 26(1):289--306

    2023

  36. [44]

    Machine Learning 54(3):187--193

    Giraud-Carrier C, Vilalta R, Brazdil P (2004) Introduction to the special issue on meta-learning. Machine Learning 54(3):187--193

    2004

  37. [45]

    Expert Systems with Applications 73:220--239

    Haixiang G, Yijing L, Shang J, et al (2017) Learning from class-imbalanced data: Review of methods and applications. Expert Systems with Applications 73:220--239

    2017

  38. [46]

    Springer

    Hastie T, Tibshirani R, Friedman JH, et al (2009) The elements of statistical learning: data mining, inference, and prediction, vol 2. Springer

    2009

  39. [47]

    Knowledge-Based Systems 212:106622

    He X, Zhao K, Chu X (2021) Automl: A survey of the state-of-the-art. Knowledge-Based Systems 212:106622

    2021

  40. [48]

    In: Learning and Intelligent Optimization

    Hutter F, Hoos HH, Leyton-Brown K (2011) Sequential model-based optimization for general algorithm configuration. In: Learning and Intelligent Optimization. Springer Berlin Heidelberg, pp 507--523

    2011

  41. [49]

    IEEE Access 8:10262--10281

    Khan I, Zhang X, Rehman M, et al (2020) A literature survey and empirical study of meta-learning for classifier selection. IEEE Access 8:10262--10281

    2020

  42. [50]

    Pattern Recognition 41(5):1718--1731

    Ko AH, Sabourin R, Britto Jr AS (2008) From dynamic classifier selection to dynamic ensemble selection. Pattern Recognition 41(5):1718--1731

    2008

  43. [51]

    In: Proceedings of the 14th International Conference on Machine Learning, vol 97

    Kubat M, Matwin S, et al (1997) Addressing the curse of imbalanced training sets: one-sided selection. In: Proceedings of the 14th International Conference on Machine Learning, vol 97. Morgan Kaufmann Publishers Inc., pp 179--186

    1997

  44. [52]

    In: Proceedings of the 7th ICML Workshop on Automated Machine Learning (AutoML), p 24

    LeDell E, Poirier S (2020) H2O A uto ML : Scalable automatic machine learning. In: Proceedings of the 7th ICML Workshop on Automated Machine Learning (AutoML), p 24

    2020

  45. [53]

    Computational Statistics 14(2):277--292

    Lee SS (1999) Regularization in skewed binary classification. Computational Statistics 14(2):277--292

    1999

  46. [54]

    Computational Statistics & Data Analysis 34(2):165--191

    Lee SS (2000) Noisy replication in skewed binary classification. Computational Statistics & Data Analysis 34(2):165--191

    2000

  47. [55]

    IEEE Transactions on Knowledge and Data Engineering 27(2):354--367

    Leyva E, Gonz \'a lez A, Perez R (2014) A set of complexity measures designed for applying meta-learning to instance selection. IEEE Transactions on Knowledge and Data Engineering 27(2):354--367

    2014

  48. [56]

    Machine Learning 107(1):209--246

    Lorena AC, Maciel AI, de Miranda PB, et al (2018) Data complexity meta-features for regression problems. Machine Learning 107(1):209--246

    2018

  49. [57]

    ACM Computing Surveys 52(5):1--34

    Lorena AC, Garcia LP, Lehmann J, et al (2019) How complex is your classification problem? a survey on measuring classification complexity. ACM Computing Surveys 52(5):1--34

    2019

  50. [58]

    Machine Learning 112(4):1131--1170

    Mohr F, Wever M (2023) Naive automated machine learning. Machine Learning 112(4):1131--1170

    2023

  51. [59]

    Expert Systems with Applications 178:115011

    Moniz N, Cerqueira V (2021) Automated imbalanced classification via meta-learning. Expert Systems with Applications 178:115011

    2021

  52. [60]

    In: International Symposium on Intelligent Data Analysis, Springer, pp 215--226

    Moniz N, Torgo L, Rodrigues F (2014) Resampling approaches to improve news importance prediction. In: International Symposium on Intelligent Data Analysis, Springer, pp 215--226

    2014

  53. [61]

    International Journal of Data Science and Analytics 3(3):161--181

    Moniz N, Branco P, Torgo L (2017 a ) Resampling strategies for imbalanced time series forecasting. International Journal of Data Science and Analytics 3(3):161--181

    2017

  54. [62]

    In: Proceedings of the First International Workshop on Learning with Imbalanced Domains: Theory and Applications, pp 129--140

    Moniz NM, Branco PO, Torgo L (2017 b ) Evaluation of ensemble methods in imbalanced regression tasks. In: Proceedings of the First International Workshop on Learning with Imbalanced Domains: Theory and Applications, pp 129--140

    2017

  55. [63]

    Knowledge and Information Systems 51(3):1067--1090

    Mor \'a n-Fern \'a ndez L, Bol \'o n-Canedo V, Alonso-Betanzos A (2017) Can classification performance be predicted by complexity measures? a study using microarray data. Knowledge and Information Systems 51(3):1067--1090

    2017

  56. [64]

    In: Proceedings of the Genetic and Evolutionary Computation Conference 2016

    Olson RS, Bartley N, Urbanowicz RJ, et al (2016) Evaluation of a tree-based pipeline optimization tool for automating data science. In: Proceedings of the Genetic and Evolutionary Computation Conference 2016. ACM, pp 485--492

    2016

  57. [65]

    In: Proceedings of the 17th International Conference on Machine Learning

    Pfahringer B, Bensusan H, Giraud-Carrier CG (2000) Meta-learning by landmarking various learning algorithms. In: Proceedings of the 17th International Conference on Machine Learning. Morgan Kaufmann Publishers Inc., pp 743--750

    2000

  58. [66]

    Knowledge-Based Systems 119:232--256

    Rathore SS, Kumar S (2017 a ) Linear and non-linear heterogeneous ensemble methods to predict the number of faults in software systems. Knowledge-Based Systems 119:232--256

    2017

  59. [67]

    Expert Systems with Applications 82:357--382

    Rathore SS, Kumar S (2017 b ) Towards an ensemble based system for predicting the number of software faults. Expert Systems with Applications 82:357--382

    2017

  60. [68]

    Pattern Analysis and Applications 17:83--96

    Reif M, Shafait F, Goldstein M, et al (2014) Automatic classifier selection for non-experts. Pattern Analysis and Applications 17:83--96

    2014

  61. [69]

    PhD thesis, Dep

    Ribeiro R (2011) Utility-based regression. PhD thesis, Dep. Computer Science, Faculty of Sciences-University of Porto

    2011

  62. [70]

    Machine Learning 109(9):1803--1835

    Ribeiro RP, Moniz N (2020) Imbalanced regression and extreme value prediction. Machine Learning 109(9):1803--1835

    2020

  63. [71]

    Knowledge-Based Systems 240:108101

    Rivolli A, Garcia LP, Soares C, et al (2022) Meta-features for meta-learning. Knowledge-Based Systems 240:108101

    2022

  64. [72]

    Information Sciences 565:262--277

    Rossi ALD, Soares C, de Souza BF, et al (2021) Micro-metastream: algorithm selection for time-changing data. Information Sciences 565:262--277

    2021

  65. [73]

    Wiley Online Library, Hoboken

    Simon D (2013) Evolutionary optimization algorithms. Wiley Online Library, Hoboken

    2013

  66. [74]

    In: Proceedings of the 26th International Conference on Neural Information Processing Systems

    Snoek J, Larochelle H, Adams RP (2012) Practical bayesian optimization of machine learning algorithms. In: Proceedings of the 26th International Conference on Neural Information Processing Systems. Curran Associates Inc., pp 2951--2959

    2012

  67. [75]

    In: Fourth International Workshop on Learning with Imbalanced Domains: Theory and Applications, PMLR, pp 38--52

    Song XY, Dao N, Branco P (2022) Distsmogn: Distributed smogn for imbalanced regression problems. In: Fourth International Workshop on Learning with Imbalanced Domains: Theory and Applications, PMLR, pp 38--52

    2022

  68. [76]

    Neurocomputing 194:45--55

    Sousa AF, Prud \^e ncio RB, Ludermir TB, et al (2016) Active learning and data manipulation techniques for generating training examples in meta-learning. Neurocomputing 194:45--55

    2016

  69. [77]

    In: Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp 847--855

    Thornton C, Hutter F, Hoos HH, et al (2013) Auto-weka: Combined selection and hyperparameter optimization of classification algorithms. In: Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp 847--855

    2013

  70. [78]

    In: Discovery Science

    Torgo L, Ribeiro R (2009) Precision and recall for regression. In: Discovery Science. Springer Berlin Heidelberg, pp 332--346

    2009

  71. [79]

    In: Portuguese Conference on Artificial Intelligence, Springer, pp 378--389

    Torgo L, Ribeiro RP, Pfahringer B, et al (2013) Smote for regression. In: Portuguese Conference on Artificial Intelligence, Springer, pp 378--389

    2013

  72. [80]

    Addison-Wesley, Reading, Mass

    Tukey J (1970) Exploratory Data Analysis. Addison-Wesley, Reading, Mass

    1970

  73. [81]

    arXiv preprint ArXiv:2109.01528

    Vakhrushev A, Ryzhkov A, Savchenko M, et al (2021) Lightautoml: Automl solution for a large financial services ecosystem. arXiv preprint ArXiv:2109.01528

    2021 arXiv

  74. [82]

    Automated machine learning: methods, systems, challenges pp 35--61

    Vanschoren J (2019) Meta-learning. Automated machine learning: methods, systems, challenges pp 35--61

    2019

  75. [83]

    ACM SIGKDD Explorations Newsletter 15(2):49--60

    Vanschoren J, Van Rijn JN, Bischl B, et al (2014) Openml: networked science in machine learning. ACM SIGKDD Explorations Newsletter 15(2):49--60

    2014

  76. [84]

    In: Proceedings of Machine Learning and Systems, pp 434--447

    Wang C, Wu Q, Weimer M, et al (2021) Flaml: A fast and lightweight automl library. In: Proceedings of Machine Learning and Systems, pp 434--447

    2021

  77. [85]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, pp 10347--10354

    Wu Q, Wang C, Huang S (2021) Frugal optimization for cost-related hyperparameters. In: Proceedings of the AAAI Conference on Artificial Intelligence, pp 10347--10354

    2021

  78. [86]

    In: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, Springer, pp 645--648

    Wu W, Kunz N, Branco P (2022) Imbalancedlearningregression-a python package to tackle the imbalanced regression problem. In: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, Springer, pp 645--648

    2022

  79. [87]

    arXiv preprint ArXiv:1810.13306v2

    Yao Q, Wang M, Chen Y, et al (2018) Taking human out of learning applications: A survey on automated machine learning. arXiv preprint ArXiv:1810.13306v2

    2018 arXiv

  80. [88]

    Journal of Artificial Intelligence Research 70:409--472

    Z \"o ller MA, Huber MF (2021) Benchmark and survey of automated machine learning frameworks. Journal of Artificial Intelligence Research 70:409--472

    2021

  81. [89]

    , " * write output.state after.block = add.period write newline

    ENTRY address archive author booktitle chapter doi edition editor eid eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year archivePrefix primaryClass adsurl adsnote version label extra.labe...

  82. [90]

    write newline

    " write newline "" before.all 'output.state := FUNCTION add.period duplicate empty 'skip "." * add.blank if FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap dupl...

  83. [91]

    write newline

    " write newline "" before.all 'output.state := FUNCTION output.doi doi empty skip "doi:" doi * "" * output if FUNCTION format.archive archivePrefix empty "" archivePrefix ":" * if FUNCTION format.primaryClass primaryClass empty "" " [" primaryClass * "] " * if FUNCTION format....

  84. [92]

    write newline

    " write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTI...

  85. [93]

    write newline

    " write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTI...

  86. [94]

    , " * write output.state after.block = add.period write newline

    ENTRY address archive author booktitle chapter edition editor eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url doi volume year archivePrefix primaryClass eid adsurl adsnote version label INTEGERS o...

  87. [95]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  88. [96]

    Available from:

    ENTRY address assignee author booktitle chapter cartographer day edition editor howpublished institution inventor journal key keywords month note number organization pages part publisher school series title type volume word year eprint doi url lastchecked updated archive archi...

  89. [97]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.