Pith. sign in

REVIEW 4 major objections 5 minor 36 references

RocketStack: Level-aware Deep Recursive Ensemble Learning Architecture

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Across 33 tabular datasets, a ten-level recursive stacking ensemble keeps gaining accuracy with depth.

desk verdict A plausible but under-specified deep-stacking configuration study; the depth-gain claims hinge on an OOF protocol that the paper never actually describes. read the letter →

arxiv 2506.16965 v3 pith:NVN7NJKG submitted 2025-06-20 cs.LG stat.ML

Çağatay Demirel This is my paper · ORCID
classification cs.LGstat.ML
keywords stackensembledeepstackingrecursivelearningout-of-foldpruningstochasticperturbationfeaturecompressionattentionmechanismtabularclassification
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 tries to establish that stacking, combining many classifiers through a meta-model, does not have to stop at one or two levels. It introduces RocketStack, a recursive stacking architecture that runs to ten levels by pruning weak learners with out-of-fold scores, perturbing those scores with Gaussian noise to avoid premature convergence, and compressing the growing feature matrix at fixed checkpoints. Across 33 datasets, the authors report that accuracy rises with depth, that the best meta-model at each level increasingly beats the best standalone ensemble, and that at the deepest level RocketStack slightly edges established deep tabular baselines while keeping runtime growth sublinear. If true, this makes deep recursive stacking a practical, scalable option for tabular machine learning rather than a curiosity.

What carries the argument

The load-bearing mechanism is a level-aware recursive stacking loop. At level 1, out-of-fold (OOF) probabilities from a pool of 20 binary or 14 multi-class classifiers are concatenated with the original features; at each later level, new OOF probabilities from the surviving models are blended with the previous level's feature matrix, optionally compressed by SFE, autoencoder, or attention-based selection at levels 3, 6, and 9, and then used to train the next meta-models. Before propagation, models are pruned by keeping only those whose OOF scores exceed an adaptive percentile threshold, optionally after adding Gaussian noise at scale lambda times the score range. A final stack-of-stacking layer concatenates selected outputs from all levels for one last estimator. This loop is what lets the architecture trade model-pool size and feature width for depth.

What would settle it

On any of the 33 datasets, instrument the pipeline so the test fold used for evaluation at level l is excluded from all training and out-of-fold generation at level l+1; if the level-10 advantage over level-1 disappears under this strict nested holdout, the depth trend is a leakage artifact.

Watch

Extended reading notes

Core claim

RocketStack's central claim is that stacking can be pushed to depth 10 with controlled pruning and periodic feature compression, and that this yields accuracy gains over shallower stacking and over the best single base learner. Across 23 binary and 10 multi-class datasets, linear mixed-effects trend tests find statistically significant positive accuracy trends with depth for the uncompressed baseline and for periodic SFE in binary and periodic attention in multi-class settings, while per-level compression and autoencoder variants fail to show consistent trends. Light Gaussian perturbation of out-of-fold scores before pruning (lambda = 0.05) improves stability and late-level gains relative to deterministic pruning, and periodic compression at levels 3, 6, and 9 gives a better accuracy-runtime trade-off than compressing at every level. The best meta-model per level increasingly outperforms the best standalone ensemble, and at level 10 untuned RocketStack reaches 88.90 percent accuracy on the binary benchmark and 94.82 percent on the multi-class benchmark, slightly above hyperparameter-tuned TabNet and Deep Forest.

Load-bearing premise

The entire depth-gain result depends on the level-wise out-of-fold predictions being computed with a clean train/test split that the pseudocode never spells out, so if a test fold is reused to build the next level's features, the reported gains could be artifacts.

Editorial extensions

If this is right

  • Stacking can be extended well beyond one or two meta-layers; on these 33 datasets the best meta-model at each level keeps improving through level 10 rather than saturating early.
  • Applying feature compression at fixed checkpoints rather than every level preserves accuracy while controlling feature count and runtime, with periodic SFE in binary and periodic attention in multi-class being the strongest compressed configurations.
  • Perturbing out-of-fold scores with light Gaussian noise before pruning improves late-level accuracy over deterministic pruning, suggesting that controlled randomness in model selection acts as a regularizer.
  • Hyperparameter optimization of base learners gives an early lead that deep recursion erases; untuned RocketStack at level 10 matches or slightly exceeds HPO-tuned versions and outperforms HPO-tuned TabNet and Deep Forest.
  • With pruning monotonically shrinking the model pool and compression bounding feature width, runtime growth with depth can remain sublinear under RocketStack's controls.

Reading between the lines

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

  • If clean out-of-fold separation holds, the level-wise accuracy curves suggest RocketStack's meta-features behave like a learned feature hierarchy; a testable extension is whether features selected at deep levels transfer across datasets.
  • The pruning-noise result mirrors Dropout for model selection; a natural next experiment is a noise schedule that anneals lambda with depth or selection by Thompson sampling over model scores, to see whether hyper-deep stacking beyond ten levels remains stable.
  • The advantage of periodic over per-level compression suggests a general accumulate-then-compress design pattern for layered ensembles that could be tested in other recursive pipelines.
  • If base-level HPO gains truly vanish by depth 10, compute budgets for deep ensembles should shift from tuning base learners toward increasing depth and pruning diversity.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The manuscript proposes RocketStack, a recursive stacking architecture that iteratively builds meta-features from out-of-fold predictions of the previous level's model pool, with pruning based on OOF scores (optionally noise-perturbed) and optional feature compression (SFE, autoencoder, attention) applied periodically or at every level. The authors evaluate the architecture on 33 OpenML datasets up to depth 10, reporting positive accuracy trends via linear mixed models, favorable runtime and dimensionality scaling, and a final-level accuracy slightly above TabNet and Deep Forest.

Significance. If the evaluation protocol is sound, the paper is a useful empirical contribution: it demonstrates that deep recursive stacking up to depth 10 is feasible with controlled feature growth, and the ablation of pruning noise and compression frequency is informative. The breadth of the study (33 datasets, multiple compression mechanisms, pruning ablations, runtime and dimensionality tracking) is a strength. The paper does not provide code, and the combination of the underspecified OOF generation and the oracle-style model selection currently prevents verification of the central quantitative claims.

major comments (4)
  1. [Section 3.3 and Algorithm 1] The construction of the level-l features P^(l) is not specified sufficiently to rule out leakage. In Algorithm 1, the line 'obtain OOF probabilities P^(l) for M^(l−1) on X^(l−1)' appears inside the outer-fold loop, but M^(l−1) contains models that, at the previous level, were trained on X^(l−1); predicting X^(l−1) with those trained models is in-sample. To produce genuinely out-of-sample probabilities, one must run an inner cross-validation that retrains the surviving models on inner folds of X^(l−1) and predicts the inner validation folds, and the pseudocode must describe how this inner CV is nested inside the outer fold and how the outer test set X_te remains untouched. Section 3.3's statement that 'every stacking level starts with OOF probabilities derived from a 5-fold cross-validation loop' is ambiguous because it is written from the perspective of the base learners at level 1. Because every headline result (LMM trends in Table A4, Figure 4, and the best-model comparisons in Table A5/Figure 5) depends on these features being OOF, this gap must be resolved before the empirical claims can be evaluated.
  2. [Section 4.2, Table A5, Figure 5] The 'best meta-model per level' accuracy is obtained by selecting, for each level, the single classifier with the highest test-set accuracy after averaging over folds and datasets. This is an oracle estimate: it uses the test labels to choose the model, so it does not represent the accuracy of any fixed algorithm that a practitioner could deploy, and it inflates the apparent gain from depth. The comparisons against the 'best standalone ensemble' (also selected on test) are therefore not a valid head-to-head benchmark. The authors should either report the accuracy of a predetermined model (e.g., the average of all surviving models, or a model chosen by an inner validation split) or explicitly label these curves as oracle upper bounds and soften the claims accordingly.
  3. [Section 4.5, Table 4] The reported comparison against TabNet and Deep Forest is under-specified. The reader cannot tell how many datasets and folds were used, whether the same train/test splits were used for all methods, how the Bayesian HPO was configured for these baselines, or whether the RocketStack numbers in Table 4 are the averaged accuracy of all level-10 meta-models (as suggested by the match with Table A2) or some other estimator. With margins of 0.51 and 1.29 percentage points and no uncertainty estimates, the claim of 'slightly surpasses' is not yet supported.
  4. [Section 3.4 (Eqs. 6-7) and Section 5.1] The conclusions about pruning noise rely on a small set of hand-picked constants (λ in {0.05, 0.1}, the 5 + 80·std(·)^2 percentile offset, the 75th-percentile attention threshold, and the compression ratio of 1/3) plus a post-hoc selection of the best configuration on the same 33 datasets. No sensitivity analysis is provided, so it is unclear whether the 'light randomization is best' finding is robust or an artifact of these particular values and of the exploratory comparison across 15 variants. At minimum, the authors should report the number of configurations examined and apply a multiple-comparisons correction across the variant selection step, or present the results as exploratory rather than confirmatory.
minor comments (5)
  1. [Author footnote] The word 'Correspondance' should be 'Correspondence'.
  2. [Tables 2 and 3] Table 2 reports accuracy in percent while Table 3 reports confidence intervals in decimal proportions; please make the units consistent.
  3. [Section 4.1.2] The SFE each-level multi-class accuracy drops sharply from 93.45% to 85.52%; consider discussing whether this is due to feature collapse or another identifiable cause.
  4. [Figure 4A] The broken y-axis in some multi-class panels makes the trends appear steeper than they are; a consistent scale would allow easier comparison across variants.
  5. [Section 4.4 and Table A6b] The feature counts are averages over datasets and therefore non-integer; this is acceptable but should be stated explicitly in the table caption or the surrounding text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: RocketStack is an empirical architecture study whose claims are measured performance comparisons, not predictions derived from fitted inputs.

full rationale

No circularity identified. RocketStack is an empirical architecture paper; the central claims are measured accuracies, linear mixed-model trend estimates, runtime profiles, and feature/model counts obtained on 33 OpenML datasets, rather than quantities that are mathematically derived from a fitted parameter and then renamed as a prediction. Algorithm 1's use of out-of-fold probabilities to construct meta-features is the standard definition of stacking, and the fact that level-(l-1) models contribute to level-l features is the architecture's intended mechanism, not a self-referential reduction, provided the OOF protocol is honestly out-of-fold. The skeptical concern that the nested cross-validation needed to keep level-l features out-of-sample is underspecified is a correctness or leakage risk that requires code inspection to resolve; it is not a circularity step under the stated rules because the paper does not define the target result in terms of its inputs. The only self-citations (Demirel et al., 2023) are prior-work context and a reference for Bayesian HPO and OOF-based pruning; they are not load-bearing justifications of the central empirical claims. External baselines such as TabNet and Deep Forest are independently implemented benchmarks, and the exploratory selection of best-performing variants on the same datasets is an in-sample model-selection concern, not a derivation that is circular by construction. No renamed known result or author-imported uniqueness theorem appears. Therefore the appropriate finding is no significant circularity, score 0.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claims rest on a set of hand-chosen hyperparameters (pruning threshold constants, noise scales, compression ratio, retention percentile, compression schedule, minimum model count) and on protocol assumptions about leakage-free OOF generation and the validity of post-hoc best-model selection. No new physical or theoretical entities are introduced.

free parameters (6)
  • pruning percentile offset constants = percentile = 5 + 80 * std(a)^2
    Hand-chosen adaptive threshold in Eq. (7); controls how many models survive each level and is not derived from theory.
  • Gaussian perturbation scales lambda = 0.05 and 0.10
    Explored values in Eq. (6); 0.05 is later reported as best, making it a selected hyperparameter rather than a prediction.
  • autoencoder compression ratio = 1/3
    Latent width set to one-third of the input in Section 3.3.2; arbitrary and applied uniformly across datasets.
  • attention retention percentile = 75th percentile, keep top 25%
    Eq. (5) retains features above Q75; chosen by hand with no sensitivity analysis.
  • periodic compression checkpoints = levels 3, 6, 9
    Selection schedule fixed by design; no evidence given that these checkpoints are better than other schedules.
  • minimum retained models t_min = not specified
    Algorithm 1 halts recursion if retained models are below t_min, but the value is never given in the paper.
assumptions (4)
  • domain assumption OOF probabilities at every level are computed with clean train/test separation via nested cross-validation.
    Section 3.3 and Algorithm 1 do not specify how level-2 and higher OOF probabilities are generated; if they reuse test folds, the accuracy trajectories are inflated.
  • domain assumption Averaging accuracy over datasets and folds and fitting a linear mixed model with classifier random intercepts is a valid trend test.
    Section 3.6; the aggregation discards dataset-level variance and the linear term is dominated by the level-0 to level-1 jump.
  • domain assumption The 33 OpenML datasets form a representative benchmark for tabular classification.
    Section 3.1; no power analysis and no external validation beyond these datasets.
  • ad hoc to paper Post-hoc selection of the best model per level is an acceptable estimator of ensemble performance.
    Section 4.2 selects the best-performing classifier at each level from the same test results used for reporting, which can inflate apparent gains.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RocketStack: Level-aware Deep Recursive Ensemble Learning Architecture." pith.science (2026). https://pith.science/paper/NVN7NJKG

@misc{pith2026250616965,
  author       = {Pith},
  title        = {Pith review of: RocketStack: Level-aware Deep Recursive Ensemble Learning Architecture},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NVN7NJKG}},
  note         = {Machine review of arXiv:2506.16965}
}
read the original abstract

Ensemble learning remains a cornerstone of machine learning, with stacking used to integrate predictions from multiple base learners through a meta-model. However, deep stacking remains uncommon due to feature redundancy, complexity, and computational burden. To address these limitations, RocketStack is introduced as a level-aware recursive stacking architecture explored up to ten stacking levels, extending beyond prior architectures. At level 1, base-learner predictions are fused with original features; at later levels, weaker learners are incrementally pruned using out-of-fold (OOF) scores. To curb early saturation, pruning is regularized by applying Gaussian perturbations at two noise scales to OOF scores prior to model selection for next-level stacking, alongside deterministic pruning. To control feature growth, periodic compression is applied at levels 3, 6, and 9 using Simple, Fast, Efficient (SFE) filtering, attention-based selection, and autoencoders. Across 33 datasets (23 binary, 10 multi-class), increasing accuracy with depth is confirmed by linear mixed-effects trend tests, and the best meta-model per level increasingly outperforms the best standalone ensemble. OOF-perturbed pruning is found to improve stability and late-level gains, while periodic compression is found to yield substantial runtime and dimensionality reductions with minimal accuracy drop. At the deepest level, accuracy slightly surpasses established deep tabular baselines. When hyperparameter optimization is performed on baseline models, early performance is boosted; however, untuned RocketStack closes the gap with depth and remains competitive at later levels. It achieves deep recursive stacking with sublinear computational growth and provides a modular, depth-aware foundation for scalable decision fusion as model pools and feature spaces evolve.

Figures

Figures reproduced from arXiv: 2506.16965 by the authors.

Figure 1
Figure 1. Schematic overview of the study thresholds, whereas dynamic schemes use validation feedback to decide which learners to retain Shen et al. (2023). Recent work has further framed pruning as an explicit optimization problem Wu et al. (2024), for example by leveraging focal diversity to adaptively trim hierarchical structures, collectively underscoring that pruning is most effective when treated as an integral design e… view at source ↗
Figure 2
Figure 2. Multi-level recursive ensemble learning (RocketStack) pipeline 8 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Estimation performance across ensembling depths for different feature selection strategies in recursive blend ensembling. Each violin plot shows accuracy distributions across datasets for a given strategy, with the thick black dashed line marking the mean and thin gray lines tracking individual trends. A: Compares overall accuracy (%) across strategies, including a no-feature-selection baseline, for binary (A1) and … view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Trend analysis results across ensembling depth. A: Shows accuracy (%) trends from baseline (individual) model performance to recursive stacking levels (L1–L10), including stack-of-stacking as the final level. Averages are computed across all surviving models, datasets,…
Figure 5
Figure 5. Figure 5: Performance comparison of RocketStack across ensembling levels against top-performing baseline ensemble models in binary (A) and multi-class (B) classification tasks. Each violin plot represents the distribution of accuracy scores across different ensembling levels, wi…
Figure 6
Figure 6. Figure 6: Elapsed time progression from baseline classification (level 0) to level 10 across RocketStack configurations for binary (A) and multi-class (B) settings. Each individual subfigure is independently normalized to the [0–1] range. C: Presents a single globally normalized…
Figure 7
Figure 7. Figure 7: Feature and model count progression across RocketStack ensembling levels. A: Shows the number of features retained from baseline to level 10 in binary (A1) and multi-class (A2) settings under various feature selection strategies. In the absence of feature selection, th…
Figure 8
Figure 8. Figure 8: Influence of base-level parameter optimization on selected RocketStack setting (incorporating OOF-light ran￾domized model pruning paired with periodic SFE for binary classifications, and periodic Attention feature selection for multi-class classifications).(A) Distribu…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 22 canonical work pages

  1. [1]

    doi:10.48185/smhs.v1i2

    work page doi:10.48185/smhs.v1i2

  2. [3]

    PeerJ Computer Science 10, e2254

    Dynamic stacking ensemble for cross-language code smell detection. PeerJ Computer Science 10, e2254. doi:10.7717/peerj-cs.2254. Alsaffar, A., Nouri-Baygi, M., Zolbanin, H., 2024a. Shielding networks: enhancing intrusion detection with hybrid feature selection and stack ensemble learning. Journal of Big Data 11, 1–24. URL: https://link.springer.com/article...

    work page doi:10.7717/peerj-cs.2254

  3. [6]

    Website fingerprinting on early QUIC traffic

    Kaggle competitions as a testbed for automated machine learning. arXiv preprint arXiv:2101.11871 . Breiman, L., 1996a. Bagging predictors. Machine Learning 24, 123–140. doi:10.1007/BF00058655. Breiman, L., 1996b. Stacked regressions. Machine Learning 24, 49–64. doi:10.1007/BF00117832. Chen, Y., Song, Q., Hu, X.,

    work page Pith review arXiv

  4. [7]

    ACM Transactions on Intelligent Systems and Technology 12, 1–36

    Techniques for automated machine learning. ACM Transactions on Intelligent Systems and Technology 12, 1–36. doi:10.1145/3447556.3447567. Choudhury, A., Mondal, A., Sarkar, S.,

    arXiv

  5. [8]

    The European Physical Journal Special Topics URL:https://link.springer.com/ article/10.1140/epjs/s11734-024-01308-x

    Scenario classification at the lhc using decision tree-based machine learning algorithms: A comparative study and review of random forest, adaboost, xgboost and lightgbm. The European Physical Journal Special Topics URL:https://link.springer.com/ article/10.1140/epjs/s11734-024-01308-x. Demirel, Tokuç, A.A., Tekin, A.T.,

    work page doi:10.1140/epjs/s11734-024-01308-x

  6. [9]

    Intelligent Systems with Applications 17, 200185

    Click prediction boosting via bayesian hyperparameter optimization-based ensemble learning pipelines. Intelligent Systems with Applications 17, 200185. doi:10.1016/j.iswa.2023.200185. Du, K., Zhang, R., Jiang, B., Zeng, J., Lu, J.,

    work page arXiv 2023

  7. [11]

    A comparison of automl tools for machine learning, deep learning and xgboost, in: 2021 International Joint Conference on Neural Networks (IJCNN), IEEE. pp. 1–8. doi:10.1109/IJCNN52387.2021.9534091. Freitas, A.A.,

    arXiv 2021

  8. [17]

    18932–18943

    Revisiting deep learn- ing models for tabular data, in: Advances in Neural Information Processing Systems, pp. 18932–18943. URL:https://proceedings.neurips.cc/paper_files/paper/2021/file/ 9d86d83f925f2149e9edb0ac3b49229c-Paper.pdf. Grubb, A., Bagnell, J.A.,

    work page 2021

Show all 36 references
  1. [18]

    Expert Systems with Applications 238, 121729

    A multilayer stacking method base on rfe-shap feature selection strategy for recognition of driver’s mental load and emotional state. Expert Systems with Applications 238, 121729. doi:10.1016/j.eswa.2023.121729. Idris, N.F., Ismail, M.A., Jaya, M.I.M., Ibrahim, A.O., Abulfaraj...

    2023

  2. [19]

    PLOS ONE 19, 1–18

    Stacking with recursive feature elimination-isolation forest for classification of diabetes mellitus. PLOS ONE 19, 1–18. doi:10.1371/journal.pone.0302595. Kumar, M., Singhal, S., Shekhar, S., Sharma, B.,

    doi

  3. [20]

    Sustainability 14, 13998

    Optimized stacking ensemble learning model for breast cancer detection and classification using machine learning. Sustainability 14, 13998. URL: https://www.mdpi.com/2071-1050/14/21/13998, doi:10.3390/su142113998. Kumar, S., Sharma, A., Shokeen, V., Azar, A., Amin, S.,

    doi

  4. [21]

    Scientific Reports URL:https://www.nature

    Meta-learning for real-world class incremental learning: a transformer-based approach. Scientific Reports URL:https://www.nature. com/articles/s41598-024-71125-8, doi:10.1038/s41598-024-71125-8. Love, J., Selker, R., Marsman, M., Jamil, T., Dropmann, D., Verhagen, J., Ly, A., ...

    doi

  5. [22]

    Journal of Statistical Software 88, 1–17

    Jasp: Graphical statistical software for common statistical designs. Journal of Statistical Software 88, 1–17. URL:https://www.jstatsoft.org/index.php/ jss/article/view/v088i02, doi:10.18637/jss.v088.i02. Mienye, I.D., Sun, Y.,

    doi

  6. [24]

    Expert Systems with Applications 211, 118611

    Dynamic ensemble pruning algorithms fusing meta-learning with heuristic parameter optimization for time series prediction. Expert Systems with Applications 211, 118611. doi:10.1016/j.eswa.2023.118611. Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R.,

    2023

  7. [26]

    SIAM/ASA Journal on Uncertainty Quantification 10, 1326–1353

    Stacking designs: Designing multifidelity computer experiments with target predictive accuracy. SIAM/ASA Journal on Uncertainty Quantification 10, 1326–1353. doi:10.1137/22M1532007. Tschalzev, A., Marton, S., Lüdtke, S., Bartelt, C.,

    doi

  8. [27]

    arXiv preprint arXiv:2407.02112 URL:https://arxiv.org/pdf/ 2407.02112

    A data-centric perspective on evaluating machine learning models for tabular data. arXiv preprint arXiv:2407.02112 URL:https://arxiv.org/pdf/ 2407.02112. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L.u., Polosukhin, I.,

    arXiv

  9. [28]

    (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc

    Attention is all you need, in: Guyon, I., Luxburg, U.V., Bengio, S., Wallach, H., Fer- gus, R., Vishwanathan, S., Garnett, R. (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc. URL:https://proceedings.neurips.cc/paper_files/paper/2017/file/ 3f5e...

    2017

  10. [29]

    Evolving Systems 12, 217–223

    Automatic tuning of hyperparameters using bayesian opti- mization. Evolving Systems 12, 217–223. URL:https://doi.org/10.1007/s12530-020-09345-2, doi:10.1007/s12530-020-09345-2. Wang, H., Tan, Z., Liang, Y., Li, F., Zhang, Z., Ju, L.,

    doi

  11. [30]

    Energy 279, 129409

    A novel multi-layer stacking ensemble wind power prediction model under tensorflow deep learning framework considering feature enhancement and data hierarchy processing. Energy 279, 129409. doi:10.1016/j.energy.2023.129409. Wolpert, D.H.,

    2023

  12. [32]

    acm.org/doi/10.1145/3633286, doi:10.1145/3633286

    URL:https://dl. acm.org/doi/10.1145/3633286, doi:10.1145/3633286. Zhao, J., Li, Y., Zhou, Z.H.,

    doi

  13. [33]

    Knowledge-Based Systems 239, 107940

    Autodes: A meta-learning framework for dynamic ensemble selection in automl. Knowledge-Based Systems 239, 107940. doi:10.1016/j.knosys.2021.107940. Zhou, Z.H.,

    2021

  14. [86]

    Zhu, J., Zhao, Z., Yin, B., Wu, C., Chen, R., Ding, Y.,

    URL:https://doi.org/10.1093/nsr/nwy108, doi:10.1093/nsr/nwy108, arXiv:https://academic.oup.com/nsr/article-pdf/6/1/74/38915135/nwy108.pdf. Zhu, J., Zhao, Z., Yin, B., Wu, C., Chen, R., Ding, Y.,

    doi

  15. [587]

    25 Ferreira, L., Pilastri, A., Martins, C.,

    URL:https://www.mdpi.com/2227-7390/ 13/4/587, doi:10.3390/math13040587. 25 Ferreira, L., Pilastri, A., Martins, C.,

    doi

  16. [1985]

    Jour- nal of Forecasting 4, 1–28

    Exponential smoothing: The state of the art. Jour- nal of Forecasting 4, 1–28. URL:https://onlinelibrary.wiley.com/doi/ abs/10.1002/for.3980040103, doi:https://doi.org/10.1002/for.3980040103, arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1002/for.3980040103. Gorishniy, Y., ...

    doi

  17. [1992]

    Neural Networks 5, 241–259

    Stacked generalization. Neural Networks 5, 241–259. doi:10.1016/S0893-6080(05) 80023-1. Wu, Y., Chow, K., Wei, W., Liu, L.,

    doi

  18. [1997]

    doi:10.1006/jcss.1997.1504

    Adecision-theoreticgeneralizationofon-linelearningandanapplication toboosting, in: JournalofComputerandSystemSciences, pp.119–139. doi:10.1006/jcss.1997.1504. Friedman, J.H.,

    1997

  19. [2001]

    Annals of Statistics 29, 1189–1232

    Greedy function approximation: A gradient boosting machine. Annals of Statistics 29, 1189–1232. doi:10.1214/aos/1013203451. Ganaie, M., Hu, M., Malik, A., Tanveer, M.,

  20. [2012]

    CRC Press

    Ensemble Methods: Foundations and Algorithms. CRC Press. doi:10.1201/b12207. Zhou, Z.H., Feng, J.,

    doi

  21. [2014]

    The journal of machine learning research 15, 1929–1958

    Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research 15, 1929–1958. Sung, C.L., Ji, Y., Mak, S., Wang, W., Tang, T.,

    1929

  22. [2017]

    (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc

    Compression-aware training of deep networks, in: Guyon, I., Luxburg, U.V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., Garnett, R. (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc. URL:https://proceedings.neurips.cc/ paper_files/pape...

    2017

  23. [2019]

    Springer, pp

    Automated machine learning for studying the trade-off between predictive accuracy and interpretability, in: Automated Machine Learning. Springer, pp. 139–152. URL:https://link. springer.com/chapter/10.1007/978-3-030-29726-8_4, doi:10.1007/978-3-030-29726-8_4. Freund, Y., Schap...

    doi

  24. [2021]

    Proceedings of the AAAI Conference on Artificial Intelligence 35, 6679–6687

    Tabnet: Attentive interpretable tabular learning. Proceedings of the AAAI Conference on Artificial Intelligence 35, 6679–6687. URL:https://doi.org/10.1609/aaai.v35i8. 16826, doi:10.1609/aaai.v35i8.16826. Baldi, P.,

    doi

  25. [2022]

    Engineering Applications of Artificial Intelligence 115, 105151

    Ensemble deep learning: A review. Engineering Applications of Artificial Intelligence 115, 105151. URL:https://www.sciencedirect.com/science/ article/pii/S095219762200269X, doi:10.1016/j.engappai.2022.105151. Gardner Jr., E.S.,

    2022

  26. [2023]

    IEEE Transactions on Evolutionary Computation 27, 1896–1911

    Sfe: A simple, fast, and efficient feature selection algorithm for high-dimensional data. IEEE Transactions on Evolutionary Computation 27, 1896–1911. doi:10.1109/TEVC.2023.3238420. Aljamaan, H.,

    1911

  27. [2024]

    An ensemble approach for multiclass classification and opti- mization of computational time, in: 2024 IEEE Recent Advances in Intelligent Computational Systems (RAICS), pp. 1–6. doi:10.1109/RAICS61201.2024.10689841. Ahadzadeh, B., Abdar, M., Safara, F., Khosravi, A., Menhaj, M...

    2024 arXiv

  28. [2025]

    Scientific Reports URL:https: //www.nature.com/articles/s41598-025-97685-x, doi:10.1038/s41598-025-97685-x

    An integrated approach of feature selection and machine learning for early detection of breast cancer. Scientific Reports URL:https: //www.nature.com/articles/s41598-025-97685-x, doi:10.1038/s41598-025-97685-x. 27 Appendix A. Additional T ables Appendix A.1. Default Model Conf...

    doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.