Pith. sign in

REVIEW 3 major objections 5 minor 47 references

QI-SMOTE, a quantum-inspired oversampling method, raises F1, G-Mean, and AUC for imbalanced MIMIC mortality prediction beyond SMOTE-family baselines—by up to 165 percent at the most severe imbalance.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

QI-SMOTE claims quantum-inspired oversampling improves imbalanced medical classification, but the described VQE step is self-defeating and the evidence is internally inconsistent.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection The VQE step in QI-SMOTE drives the evolved state orthogonal to the sample it's supposed to enrich, and the numbers don't match the tables. the 3 major comments →

arxiv 2509.02863 v1 pith:SOQFRPXJ submitted 2025-09-02 cs.LG cs.AI

Enhancing Machine Learning for Imbalanced Medical Data: A Quantum-Inspired Approach to Synthetic Oversampling (QI-SMOTE)

Vikas Kashtriya , Pardeep Singh This is my paper
classification cs.LG cs.AI
keywords quantum-inspired computingclass imbalancesynthetic oversamplingSMOTEmedical machine learningMIMIC datasetvariational quantum eigensolverdata preprocessing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 introduces QI-SMOTE, a preprocessing method that applies quantum-inspired state encoding—superposition, layered entanglement via CNOT, CZ, and Toffoli gates, then VQE-based evolution—to minority-class samples before classical SMOTE interpolation, and claims this produces synthetic medical data that better preserve feature dependencies. The authors test it on MIMIC-III and MIMIC-IV mortality prediction at imbalance ratios from 2.05 to 20, comparing with SMOTE, ADASYN, Borderline-SMOTE, SMOTE-ENN, SMOTE-TOMEK, and SVM-SMOTE across RF, SVM, LR, KNN, GB, and neural classifiers. Their headline empirical claim is that QI-SMOTE consistently raises F1, G-Mean, and AUC beyond these baselines, with the largest relative gains at the most severe imbalance (165% F1 improvement over the original dataset on MIMIC-IV(20)). The paper also argues the entanglement step prevents unrealistic physiological combinations (e.g., high blood pressure with very low heart rate) that classical SMOTE can generate. A sympathetic reader would care because, if correct, it offers a reusable, algorithm-agnostic preprocessing step that makes imbalanced clinical ML both more accurate and more plausible at the sample level.

Core claim

QI-SMOTE encodes each minority sample into an n-qubit state by mapping features to RY rotation angles after a Hadamard initialization, entangles the qubits with CNOT, CZ, and Toffoli gates, then forms the Hamiltonian H = |Ψ⟩⟨Ψ| from the resulting state vector and runs VQE to obtain an 'evolved' state. The real part of that evolved state is appended to the original feature table, and the enlarged dataset is passed to classical SMOTE for interpolation. The paper's central claim is that this pipeline yields synthetic minority samples that preserve high-order feature correlations—unlike linear interpolation in raw feature space—and that classifiers trained on QI-SMOTE-balanced data outperform th

What carries the argument

The load-bearing object is the layered-entanglement state |Ψ⟩ produced per minority sample: Hadamard + RY(f_i) encode features, CNOT/CZ/Toffoli layers couple qubits, and VQE with Hamiltonian H = |Ψ⟩⟨Ψ| is used to 'evolve' the state. The paper uses this evolved state, after taking its real part, as an enriched feature vector appended to the original data, so that SMOTE interpolates in a space the method claims preserves physiological co-dependencies rather than raw coordinate space.

Load-bearing premise

The method's gains rest on the assumption that the VQE step enriches each minority sample while keeping its information; the Hamiltonian it defines, H = |Ψ⟩⟨Ψ|, instead drives the optimized state toward the complement of the encoded sample, so the assumed preservation may not be what happens.

What would settle it

Take any minority sample used in the paper, form |Ψ⟩ by the algorithm's encoding, run the VQE routine with H = |Ψ⟩⟨Ψ|, and record |⟨Ψ|Ψ_evolved⟩|. If that overlap is near zero (as minimization of ⟨ψ|H|ψ⟩ implies), the central mechanism—evolved states that preserve and enrich the sample—is not operative, and the reported gains must come from elsewhere in the pipeline.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If QI-SMOTE works as claimed, F1, G-Mean, and AUC on imbalanced medical tabular data improve across tree, kernel, and neural classifiers whenever conventional SMOTE-family resampling is replaced by QI-SMOTE preprocessing.
  • The improvement is largest at severe imbalance: the paper reports 89.5% and 165.3% F1 improvements over the unaltered dataset on MIMIC-IV(10) and MIMIC-IV(20), respectively, where baselines plateau.
  • Because the method is a preprocessing step, it can be dropped into existing ML pipelines without modifying the classifier.
  • Entanglement co-modulates features, so synthetic samples avoid physiologically implausible combinations such as high systolic blood pressure with very low heart rate.
  • In its current classical-simulation form, QI-SMOTE costs orders of magnitude more compute and memory than SMOTE; the authors argue this tradeoff would disappear on real quantum hardware.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because minimizing ⟨ψ|H|ψ⟩ for H = |Ψ⟩⟨Ψ| pushes the optimized state toward the orthogonal complement of |Ψ⟩, a straightforward check of the inner product ⟨Ψ|Ψ_evolved⟩ on actual minority samples would show whether VQE is preserving or erasing the encoded data; this is a testable consequence the paper does not report.
  • The algorithm never specifies how a 2^n-dimensional real vector (512 components for 9 features) is appended to an n-feature table, so the empirical gains may be an artifact of SMOTE interpolating a high-dimensional expanded space rather than of the quantum-inspired correlations; a control that appends random or classically projected noise would isolate the mechanism.
  • If the gains survive that noise-control ablation, QI-SMOTE's contribution reduces to 'expand features then SMOTE', which would transfer directly to other tabular imbalance problems; if not, the quantum gates and VQE are not the active ingredient.
  • The runtime/memory gap suggests that until real quantum hardware is available, the practical value of QI-SMOTE depends on whether the F1/AUC gains justify roughly a thousand-fold slowdown relative to SMOTE; comparing at a fixed compute budget would be a more decision-relevant benchmark.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes QI-SMOTE, a preprocessing method that augments SMOTE with quantum-inspired operations. Minority samples are encoded into qubit rotations and entangled with CNOT/CZ/Toffoli gates; the resulting state |Psi> is evolved by a VQE loop whose Hamiltonian is defined as H = |Psi><Psi|; the real part of the evolved state is appended to the dataset, and SMOTE is applied on the augmented feature space. The method is evaluated on MIMIC-III and MIMIC-IV mortality prediction at imbalance ratios 2.05-20 with six classifiers and compared to SMOTE-family baselines, reporting F1, G-Mean, and AUC improvements. The central claim is that QI-SMOTE produces more informative synthetic data and outperforms conventional resampling methods on imbalanced medical data.

Significance. If the claims were supported, QI-SMOTE would be a reusable, structure-aware oversampling method for imbalanced clinical tabular data. The paper deserves credit for a fairly detailed algorithmic description, evaluation on public medical datasets, a runtime/memory benchmark, and a Wilcoxon test. However, the defining quantum-evolution step is internally inconsistent as specified, and the append operation is undefined; therefore the paper does not establish a mechanism for the reported gains. The significance of the empirical tables depends on a mechanism that the manuscript itself contradicts.

major comments (3)
  1. [3.2, Algorithm steps 2.1.1-2.1.2; Appendix A.6] The main text defines H = |Psi><Psi| (step 2.1.1) and then asks VQE to minimize <psi(theta)|H|psi(theta)> = |<psi(theta)|Psi>|^2. This rank-one projector has ground states in the orthogonal complement of |Psi>. Thus a converged VQE returns |Psi_evolved> with zero overlap with the minority sample it was supposed to preserve and enrich; even partial convergence systematically reduces that overlap. Appendix A.6 instead defines an Ising Hamiltonian H = sum w_ij Z_i Z_j + sum b_i Z_i with 'learned' weights, but gives no learning procedure or weight values. The two specifications are contradictory, and neither supports the claim that the evolution step enriches the sample. Because this VQE step is the only component that distinguishes QI-SMOTE from classical SMOTE, the reported performance advantage has no described causal basis.
  2. [3.2, step 3.1] The operation 'Append |Psi_real> to the original dataset D' is undefined. For an n-feature sample, |Psi_real> is a vector of length 2^n (512 components for the 9 features used here). The paper does not specify whether these components become new columns, how they are aligned with the original rows, how negative or complex-valued entries are handled, or how SMOTE interpolates in the resulting mixed feature space. This step is load-bearing for the 'enriched feature space' claim, and its undefined nature makes the algorithm irreproducible.
  3. [5.3] The statistical significance analysis covers only one comparison (QI-SMOTE vs. SMOTE-ENN on MIMIC-IV (10)), while the conclusion claims broad superiority over all conventional resampling methods across six dataset variants and many classifiers. No correction for multiple comparisons is reported. This single paired test does not substantiate the general claim, even setting aside the mechanism problem raised in Major Comment 1.
minor comments (5)
  1. [5.2] The text describes QI-SMOTE's improvements as 'exponential'; the reported values are ordinary percentage improvements over a low baseline and should be described as such.
  2. [Abstract and Section 4] The abstract promises evaluation with Neural Networks, but the results tables do not report a Neural Network classifier.
  3. [Equation (1)] Equation (1) is mis-typeset/torn in the manuscript; the formula for percentage improvement should be typeset cleanly.
  4. [Figures 2-3] Several figures are labeled 'Figure 2. (continued)' and 'Figure 3. (continued)' with no explanatory captions for the repeated panels; captions and numbering should be consolidated.
  5. [Section 4.2] Feature selection and preprocessing for MIMIC-III/IV are described only at a high level; no code or detailed pipeline is provided, which hinders reproducibility of the empirical tables.

Circularity Check

2 steps flagged

The central VQE evolution step is self-defeating by construction: Algorithm 2.1.1 defines H = |Ψ⟩⟨Ψ|, so minimizing the VQE expectation drives |Ψevolved⟩ into the orthogonal complement of the minority sample it was supposed to enrich.

specific steps
  1. self definitional [Algorithm steps 2.1.1–2.1.2 (Section 3.2)]
    "2.1.1. Construct the Hamiltonian matrix H using the outer product: H = |Ψ⟩⟨Ψ| 2.1.2. Use the VQE† to find the evolved state |Ψevolved⟩. VQE will minimize the expectation value of H with respect to a parameterized quantum state."

    H is the rank-one projector onto the very state |Ψ⟩ being evolved. Minimizing ⟨ψ(θ)|H|ψ(θ)⟩ = |⟨ψ(θ)|Ψ⟩|² has as ground states only states in the orthogonal complement of |Ψ⟩ (eigenvalue 0), since the projector's eigenvalues are 1 and 0. Thus a converged VQE drives |Ψevolved⟩ to have zero overlap with |Ψ⟩. The paper's central claim that quantum evolution 'preserves and enriches' the minority sample is therefore inverted by construction: the only quantum-specific transformation in the pipeline discards the sample's state rather than enriching it, and no independent information source is introduced.

  2. other [Appendix A.6 vs Algorithm 2.1.1]
    "H = Σ w_ij Z_i Z_j + Σ b_i Z_i ... Here, Z_i are Pauli-Z operators, and w_ij, b_i are learned weights reflecting the coupling between features. ... For the VQE optimization step, the following hyperparameters were used across all datasets: Optimizer: COBYLA, Maximum Iterations: 100 per sample, Convergence Tolerance: 10^-6, Ansatz Structure: 2 entangling layers using RY rotations and CZ gates"

    The main-text algorithm defines H = |Ψ⟩⟨Ψ|, while Appendix A.6 defines a different Ising Hamiltonian with 'learned' weights w_ij, b_i, but no learning procedure, loss function, or weight values are given. Since the VQE output is entirely determined by H, the algorithm is not uniquely specified: either the main-text H is used, in which case the optimized state is orthogonal to the sample (as shown above), or the appendix H is used, in which case the claimed 'learned' evolution is undefined. Thus the paper's quantum enrichment mechanism is either self-defeating or vacuous.

full rationale

The paper's quantitative comparisons against SMOTE-family baselines on MIMIC-III/IV are external and not themselves circular: the reported F1, G-Mean, and AUC numbers could in principle support the empirical claim. However, the load-bearing mechanism that distinguishes QI-SMOTE from classical SMOTE—the VQE 'quantum evolution' step—is self-referential by construction. Algorithm 2.1.1 sets H = |Ψ⟩⟨Ψ|, and VQE minimizes |⟨ψ|Ψ⟩|², whose ground state lies in the orthogonal complement of the minority sample. A converged VQE therefore produces a state orthogonal to the encoded sample, directly contradicting the paper's claim that evolution 'preserves complex data structures' and 'enriches' the representation. The separate step of appending the 2^n-dimensional |Ψ_real⟩ to an n-feature table is also undefined—for 9 features this means 512 new columns per row—but the VQE inversion alone is sufficient to invalidate the described causal chain. Appendix A.6 attempts to replace the projector with an Ising Hamiltonian with 'learned' weights, but no learning procedure is supplied, so this does not rescue the mechanism. The central claim of the paper therefore reduces to a transformation that, by the paper's own equations, discards the very information it claims to preserve. This is a definitional circularity in the core algorithm, warranting a score of 8.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 2 invented entities

The central claim rests on four unproven premises: the Qiskit simulation is faithful (no verification), the gate stack encodes physiological correlations (asserted, not demonstrated), the 2^n-to-n append is well-defined (dimensionally incoherent), and VQE with a sample-projector Hamiltonian preserves information (mathematically false; the minimum is attained in the orthogonal complement). Free parameters include VQE settings, undisclosed Hamiltonian weights, unspecified superposition/entanglement factors, and all classifier/preprocessing choices. No invented entity in the paper has independent falsifiable evidence; the |Psi_real> vector and the layered-entanglement architecture are only validated through unablated downstream metrics.

free parameters (4)
  • VQE hyperparameters (COBYLA, 100 iterations, tolerance 1e-6, 2 ansatz layers) = 2 RY+CZ layers; 100 iterations per sample
    Chosen by the authors for all datasets; no sensitivity analysis, yet they determine the quantum evolution output claimed to drive the gains (Appendix A.6).
  • Hamiltonian weights w_ij and biases b_i (Ising form)
    Appendix A.6 calls these learned weights but gives no fitting procedure or values; if fit to data they are undisclosed fitted parameters.
  • Superposition coefficients and entanglement factors
    Section 3.2 says these quantum parameters are defined at initialization but their values or setting procedure never appear in the paper.
  • Classifier hyperparameters and preprocessing choices
    No classifier configuration, categorical encoding (gender, admission_type, insurance, icd_codes), or train/test split ratio is reported; these choices determine the Table 2-3 numbers.
axioms (4)
  • domain assumption Classical simulation of the described quantum circuit (Qiskit) faithfully implements the QI-SMOTE transformation
    Invoked throughout Sections 3 and 5.5; no circuit verification or equivalence argument is provided.
  • domain assumption Feature-wise RY encoding plus CNOT/CZ/Toffoli gates models physiological co-variation
    Sections 3.2 and 5.4 assert the mechanism; no evidence links this gate sequence to preserved feature correlations.
  • ad hoc to paper Appending a 2^n-dimensional state vector |Psi_real> to an n-feature dataset is a well-defined operation for SMOTE
    Required by Algorithm steps 2.1.3-4.1; semantics never defined, dimension mismatch unresolved.
  • ad hoc to paper VQE minimization of <psi|Psi><Psi|psi> preserves minority-class information
    The stated objective is minimized by states orthogonal to |Psi>, contradicting the assumed information preservation.
invented entities (2)
  • |Psi_real> enriched synthetic state vectors no independent evidence
    purpose: Represent each minority sample after quantum evolution; appended to the dataset before SMOTE to encode feature interdependencies and improve synthetic sample quality
    The vectors have no falsifiable property outside the pipeline; their claimed structure-preserving effect is supported only by unablated downstream metrics and visual inspection.
  • Layered entanglement architecture (CNOT+CZ+Toffoli stacks) no independent evidence
    purpose: Intended to model pairwise and higher-order feature dependencies during synthetic sample generation
    Its effect is never isolated by ablation or quantified, so it functions as an asserted mechanism rather than a verified one.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Machine Learning for Imbalanced Medical Data: A Quantum-Inspired Approach to Synthetic Oversampling (QI-SMOTE)." pith.science (2026). https://pith.science/paper/SOQFRPXJ

@misc{pith2026250902863,
  author       = {Pith},
  title        = {Pith review of: Enhancing Machine Learning for Imbalanced Medical Data: A Quantum-Inspired Approach to Synthetic Oversampling (QI-SMOTE)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SOQFRPXJ}},
  note         = {Machine review of arXiv:2509.02863}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Class imbalance remains a critical challenge in machine learning (ML), particularly in the medical domain, where underrepresented minority classes lead to biased models and reduced predictive performance. This study introduces Quantum-Inspired SMOTE (QI-SMOTE), a novel data augmentation technique that enhances the performance of ML classifiers, including Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), k-Nearest Neighbors (KNN), Gradient Boosting (GB), and Neural Networks, by leveraging quantum principles such as quantum evolution and layered entanglement. Unlike conventional oversampling methods, QI-SMOTE generates synthetic instances that preserve complex data structures, improving model generalization and classification accuracy. We validate QI-SMOTE on the MIMIC-III and MIMIC-IV datasets, using mortality detection as a benchmark task due to their clinical significance and inherent class imbalance. We compare our method against traditional oversampling techniques, including Borderline-SMOTE, ADASYN, SMOTE-ENN, SMOTE-TOMEK, and SVM-SMOTE, using key performance metrics such as Accuracy, F1-score, G-Mean, and AUC-ROC. The results demonstrate that QI-SMOTE significantly improves the effectiveness of ensemble methods (RF, GB, ADA), kernel-based models (SVM), and deep learning approaches by producing more informative and balanced training data. By integrating quantum-inspired transformations into the ML pipeline, QI-SMOTE not only mitigates class imbalance but also enhances the robustness and reliability of predictive models in medical diagnostics and decision-making. This study highlights the potential of quantum-inspired resampling techniques in advancing state-of-the-art ML methodologies.

Figures

Figures reproduced from arXiv: 2509.02863 by Pardeep Singh, Vikas Kashtriya.

Figure 2
Figure 2. Figure 2: Visualization of synthetic sample distributions in a 2D feature space generated by various resampling techniques, including QI-SMOTE. Each plot highlights the spatial distribution of minority and synthetic samples relative to the majority class. The original dataset, characterized by a clear demarcation between the classes, vividly highlighted the prevalent issue of class imbalance. This visualization unde… view at source ↗
Figure 2
Figure 2. Figure 2: (continued). 5.2 Quantitative Results Analysing the F-measure data across MIMIC-III and MIMIC-IV datasets as shown in the [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Three-dimensional scatter plots illustrating the effect of QI-SMOTE and other oversampling methods in a higher-dimensional space. The QI-SMOTE-generated samples exhibit more uniform dispersion and coverage of the minority class [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 3
Figure 3. Figure 3: (continued) [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of F1 Score of each oversampling algorithm on MIMIC-III and MIMIC-IV Datasets and their imbalanced variants [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Synthetic Sample Distribution — SMOTE vs. QI-SMOTE on MIMIC-IV (10) This figure compares synthetic sample distributions generated by classical SMOTE and QI-SMOTE in a 2D PCA projection space. Green points represent original minority class samples. SMOTE-generated samples (orange) tend to extrapolate into less realistic zones of the feature space, while QI-SMOTE-generated samples (blue) remain more densely … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 32 canonical work pages · 1 internal anchor

  1. [1]

    and Ma, Y.: Imbalanced learning: foundations, algorithms, and applications

    He, H. and Ma, Y.: Imbalanced learning: foundations, algorithms, and applications. John Wiley & Sons, Inc, Hoboken, New Jersey (2013)

    work page 2013

  2. [2]

    Johnson, J. M. and Khoshgoftaar, T. M.: Survey on deep learning with class imbalance. J. Big Data 6, 27 (2019). https://doi.org/10.1186/s40537-019-0192-5

    doi:10.1186/s40537-019-0192-5 2019

  3. [3]

    V., Bowyer, K

    Chawla, N. V., Bowyer, K. W., Hall, L. O., and Kegelmeyer, W. P.: SMOTE: Synthetic Minority Over-sampling Technique. J. Artif. Intell. Res. 16, 321–357 (2002). https://doi.org/10.1613/jair.953 20

    doi:10.1613/jair.953 2002

  4. [4]

    V.: SMOTE for Learning from Imbalanced Data: Progress and Challenges, Marking the 15 -year Anniversary

    Fernandez, A., Garcia, S., Herrera, F., and Chawla, N. V.: SMOTE for Learning from Imbalanced Data: Progress and Challenges, Marking the 15 -year Anniversary. J. Artif. Intell. Res. 61, 863 –905 (2018). https://doi.org/10.1613/jair.1.11192

    work page doi:10.1613/jair.1.11192 2018

  5. [5]

    Krawczyk, B.: Learning from imbalanced data: open challenges and future directions. Prog. Artif. Intell. 5, 221 – 232 (2016). https://doi.org/10.1007/s13748-016-0094-0

    doi:10.1007/s13748-016-0094-0 2016

  6. [6]

    and Barber, D.: On -line learning from finite training sets

    Sollich, P. and Barber, D.: On -line learning from finite training sets. Europhys. Lett. 38, 477 –482 (1997). https://doi.org/10.1209/epl/i1997-00271-3

    work page doi:10.1209/epl/i1997-00271-3 1997

  7. [7]

    A., Wardoyo, R., Musdholifah, A., et al.: SMOTE for Handling Imbalanced Data Problem : A Review

    Pradipta, G. A., Wardoyo, R., Musdholifah, A., et al.: SMOTE for Handling Imbalanced Data Problem : A Review. In: 2021 Sixth International Conference on Informatics and Computing (ICIC). IEEE, Jakarta, Indonesia, pp. 1–8 (2021)

    work page 2021

  8. [8]

    and Lusa, L.: SMOTE for high -dimensional class -imbalanced data

    Blagus, R. and Lusa, L.: SMOTE for high -dimensional class -imbalanced data. BMC Bioinformatics 14, 106 (2013). https://doi.org/10.1186/1471-2105-14-106

    doi:10.1186/1471-2105-14-106 2013

  9. [9]

    Sun, Y., Wong, A. K. C., and Kamel, M. S.: Classification of imbalanced data: A review. Int. J. Pattern Recogn. Artif. Intell. 23, 687–719 (2009). https://doi.org/10.1142/S0218001409007326

    doi:10.1142/s0218001409007326 2009

  10. [10]

    P., and De Souza Britto, A.: Cost -sensitive learning for imbalanced data streams

    Loezer, L., Enembreck, F., Barddal, J. P., and De Souza Britto, A.: Cost -sensitive learning for imbalanced data streams. In: Proceedings of the 35th Annual ACM Symposium on Applied Computing. ACM, Brno, Czech Republic, pp. 498–504 (2020)

    work page 2020

  11. [11]

    A.: An introduction to matrix concentration inequalities

    Tropp, J. A.: An introduction to matrix concentration inequalities. Now, Boston Delft (2015)

    work page 2015

  12. [12]

    Nature 549, 195 –202 (2017)

    Biamonte, J., Wittek, P., Pancotti, N., et al.: Quantum ML. Nature 549, 195 –202 (2017). https://doi.org/10.1038/nature23474

    doi:10.1038/nature23474 2017

  13. [13]

    Expert Syst

    Xiao, J., Yan, Y., Zhang, J., and Tang, Y.: A quantum -inspired genetic algorithm for k-means clustering. Expert Syst. Appl. 37, 4966–4973 (2010). https://doi.org/10.1016/j.eswa.2009.12.017

    doi:10.1016/j.eswa.2009.12.017 2010

  14. [14]

    Batista, G. E. A. P. A., Prati, R. C., and Monard, M. C.: A study of the behavior of several methods for balancing ML training data. SIGKDD Explor. Newsl. 6, 20–29 (2004). https://doi.org/10.1145/1007730.1007735

    work page arXiv 2004

  15. [15]

    In: D.-S

    Han, H., Wang, W.-Y., and Mao, B.-H.: Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning. In: D.-S. Huang, X.-P. Zhang, and G.-B. Huang, eds., Advances in Intelligent Computing. Springer Berlin Heidelberg, Berlin, Heidelberg, pp. 878–887 (2005)

    work page 2005

  16. [16]

    A., and Li, S.: ADASYN: Adaptive synthetic sampling approach for imbalanced learning

    He, H., Bai, Y., Garcia, E. A., and Li, S.: ADASYN: Adaptive synthetic sampling approach for imbalanced learning. In: 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence), pp. 1322–1328 (2008)

    work page 2008

  17. [17]

    E., et al.: SMOTE –ENN-Based Data Sampling and Improved Dynamic Ensemble Selection for Imbalanced Medical Data Classification

    Lamari, M., Azizi, N., Hammami, N. E., et al.: SMOTE –ENN-Based Data Sampling and Improved Dynamic Ensemble Selection for Imbalanced Medical Data Classification. In: F. Saeed, T. Al -Hadhrami, F. Mohammed, and E. Mohammed, eds., Advances on Smart and Soft Computing. Springer, Singapore, pp. 37–49 (2021)

    work page 2021

  18. [18]

    In: 2016 IEEE International Conference of Online Analysis and Computing Science (ICOACS)

    Zeng, M., Zou, B., Wei, F., et al.: Effective prediction of three common diseases by combining SMOTE with Tomek links technique for imbalanced medical data. In: 2016 IEEE International Conference of Online Analysis and Computing Science (ICOACS). IEEE, Chongqing, China, pp. 225–228 (2016)

    work page 2016

  19. [19]

    V., and Krasser, S.: SVMs Modeling for Highly Imbalanced Classification

    Tang, Y., Zhang, Y.-Q., Chawla, N. V., and Krasser, S.: SVMs Modeling for Highly Imbalanced Classification. IEEE Trans. Syst. Man, Cybern. B 39, 281–288 (2009). https://doi.org/10.1109/TSMCB.2008.2002909

    work page arXiv 2009

  20. [20]

    Zhang, J., Wang, T., Ng, W. W. Y., et al.: Minority Oversampling Using Sensitivity. In: 2020 International Joint Conference on Neural Networks (IJCNN). IEEE, Glasgow, United Kingdom, pp. 1 –7 (2020)

    work page 2020

  21. [21]

    A., Wardoyo, R., Musdholifah, A., and Sanjaya, I

    Pradipta, G. A., Wardoyo, R., Musdholifah, A., and Sanjaya, I. N. H.: Radius -SMOTE: A New Oversampling Technique of Minority Samples Based on Radius Distance for Learning from Imbalanced Data. IEEE Access 9, 74763–74777 (2021). https://doi.org/10.1109/ACCESS.2021.3080316

    work page arXiv 2021

  22. [22]

    V.: DeepSMOTE: Fusing Deep Learning and SMOTE for Imbalanced Data

    Dablain, D., Krawczyk, B., and Chawla, N. V.: DeepSMOTE: Fusing Deep Learning and SMOTE for Imbalanced Data. IEEE Trans. Neural Netw. Learn. Syst. 34, 6390 –6404 (2023). https://doi.org/10.1109/TNNLS.2021.3136503

    work page arXiv 2023

  23. [23]

    S., Abreu, P

    Santos, M. S., Abreu, P. H., García -Laencina, P. J., et al.: A new cluster -based oversampling method for improving survival prediction of hepatocellular carcinoma patients. J. Biomed. Inform. 58, 49 –59 (2015). https://doi.org/10.1016/j.jbi.2015.09.012

    work page doi:10.1016/j.jbi.2015.09.012 2015

  24. [24]

    Xu, Z., Shen, D., Nie, T., and Kou, Y.: A hybrid sampling algorithm combining M -SMOTE and ENN based on Random Forest for medical imbalanced data. J. Biomed. Inform. 107, 103465 (2020). 21 https://doi.org/10.1016/j.jbi.2020.103465

    work page arXiv 2020

  25. [25]

    K., Reddy, V

    Rout, S., Mallick, P. K., Reddy, V. N., and Kumar, S.: A tailored particle swarm and Egyptian vulture optimization-based synthetic minority-oversampling technique for class imbalance problem. Information 13, 386 (2022). https://doi.org/10.3390/info13080386

    work page doi:10.3390/info13080386 2022

  26. [26]

    Cao, W., Yang, P., Ming, Z., et al.: An improved fuzziness based random vector functional link network for liver disease detection. In: 2020 IEEE 6th International Conference on Big Data Security on Cloud (BigDataSecurity), IEEE International Conference on High Performance and Smart Computing (HPSC), and IEEE International Conference on Intelligent Data a...

    work page 2020

  27. [27]

    In: 2021 IEEE 4th International Conference on Big Data and Artificial Intelligence (BDAI)

    Xu, B., Wang, W., Yang, R., and Han, Q.: An improved unbalanced data classification method based on hybrid sampling approach. In: 2021 IEEE 4th International Conference on Big Data and Artificial Intelligence (BDAI). IEEE, Qingdao, China, pp. 125–129 (2021)

    work page 2021

  28. [28]

    -B., Zou, C

    Wang, J. -B., Zou, C. -A., and Fu, G. -H.: AWSMOTE: An SVM -based adaptive weighted SMOTE for class - imbalance learning. Sci. Program. 2021, 1–18 (2021). https://doi.org/10.1155/2021/9947621

    doi:10.1155/2021/9947621 2021

  29. [29]

    IJRTER 3, 444 –449 (2017)

    A review on imbalanced data handling using undersampling and oversampling technique. IJRTER 3, 444 –449 (2017). https://doi.org/10.23883/IJRTER.2017.3168.0UWXM

    work page arXiv 2017

  30. [30]

    Springer International Publishing, Cham (2018)

    Fernández, A., García, S., Galar, M., et al.: Learning from imbalanced data sets. Springer International Publishing, Cham (2018)

    work page 2018

  31. [31]

    L., and Yao, X.: Resampling -based ensemble methods for online class imbalance learning

    Wang, S., Minku, L. L., and Yao, X.: Resampling -based ensemble methods for online class imbalance learning. IEEE Trans. Knowl. Data Eng. 27, 1356–1368 (2015). https://doi.org/10.1109/TKDE.2014.2345380

    work page arXiv 2015

  32. [32]

    PeerJ Computer Science 7, e832 (2021)

    Pes, B., and Lai, G.: Cost-sensitive learning strategies for high-dimensional and imbalanced data: a comparative study. PeerJ Computer Science 7, e832 (2021). https://doi.org/10.7717/peerj -cs.832

    doi:10.7717/peerj 2021

  33. [33]

    Kong, J., Kowalczyk, W., Menzel, S., and Bäck, T.: Improving imbalanced classification by anomaly detection. In: B. T. Bäck, M. Preuss, A. Deutz, et al. (eds.), Parallel Problem Solving from Nature – PPSN XVI. Springer International Publishing, Cham, pp. 512–523 (2020)

    work page 2020

  34. [34]

    D.: Quantum computing: An applied approach

    Hidary, J. D.: Quantum computing: An applied approach. Springer International Publishing, Cham (2019)

    work page 2019

  35. [35]

    Springer Berlin Heidelberg, Berlin, Heidelberg (2000)

    Bouwmeester, D., Ekert, A., and Zeilinger, A.: The physics of quantum information. Springer Berlin Heidelberg, Berlin, Heidelberg (2000)

    work page 2000

  36. [36]

    A Quantum Approach to Synthetic Minority Oversampling Technique (SMOTE)

    Mohanty, N., Behera, B. K., Ferrie, C., and Dash, P.: A quantum approach to synthetic minority oversampling technique (SMOTE). https://doi.org/10.48550/ARXIV.2402.17398 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2402.17398 2024

  37. [37]

    Esfeld, M.: Quantum entanglement and a metaphysics of relations. Stud. Hist. Phil. Sci. Part B: Stud. Hist. Philos. Modern Phys. 35, 601–617 (2004). https://doi.org/10.1016/j.shpsb.2004.04.008

    work page doi:10.1016/j.shpsb.2004.04.008 2004

  38. [38]

    P., and Loss, D.: Quantum computers and quantum coherence

    DiVincenzo, D. P., and Loss, D.: Quantum computers and quantum coherence. J. Magn. Magn. Mater. 200, 202– 218 (1999). https://doi.org/10.1016/S0304-8853(99)00315-7

    work page doi:10.1016/s0304-8853(99)00315-7 1999

  39. [39]

    P., and Jayapandian, N.: Challenges and opportunities: Quantum computing in ML

    Shubham, S. P., and Jayapandian, N.: Challenges and opportunities: Quantum computing in ML. In: 2019 Third International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I -SMAC). IEEE, Palladam, India, pp. 598–602 (2019)

    work page 2019

  40. [40]

    Johnson, A., Pollard, T., & Mark, R. (2016). MIMIC -III Clinical Database (version 1.4). PhysioNet. RRID:SCR_007345. https://doi.org/10.13026/C2XW26

    doi:10.13026/c2xw26 2016

  41. [41]

    Johnson, A. E. W., Pollard, T. J., Shen, L., et al.: MIMIC -III, a freely accessible critical care database. Sci. Data 3, 160035 (2016). https://doi.org/10.1038/sdata.2016.35

    doi:10.1038/sdata.2016.35 2016

  42. [42]

    L., Amaral, L

    Goldberger, A. L., Amaral, L. A. N., Glass, L., et al.: PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation 101, e215 (2000). https://doi.org/10.1161/01.CIR.101.23.e215

    doi:10.1161/01.cir.101.23.e215 2000

  43. [43]

    A., & Mark, R

    Johnson, A., Bulgarelli, L., Pollard, T., Gow, B., Moody, B., Horng, S., Celi, L. A., & Mark, R. (2024). MIMIC- IV (version 3.1). PhysioNet. RRID:SCR_007345. https://doi.org/10.13026/kpb9 -mt58

    work page doi:10.13026/kpb9 2024

  44. [44]

    Johnson, A. E. W., Bulgarelli, L., Shen, L., et al.: MIMIC-IV, a freely accessible electronic health record dataset. Sci. Data 10, 1 (2023). https://doi.org/10.1038/s41597-022-01899-x

    doi:10.1038/s41597-022-01899-x 2023

  45. [45]

    Röösli, E., Bozkurt, S., and Hernandez -Boussard, T.: Peeking into a black box, the fairness and generalizability of a MIMIC-III benchmarking model. Sci. Data 9, 24 (2022). https://doi.org/10.1038/s41597 -021-01110-7

    doi:10.1038/s41597 2022

  46. [46]

    Purushotham, S., Meng, C., Che, Z., and Liu, Y.: Benchmarking deep learning models on large healthcare 22 datasets. J. Biomed. Inform. 83, 112–134 (2018). https://doi.org/10.1016/j.jbi.2018.04.007

    doi:10.1016/j.jbi.2018.04.007 2018

  47. [47]

    Yang, B., Tian, G., Luttrell, J., Gong, P., Zhang, C.: A quantum-based oversampling method for classification of highly imbalanced and overlapped data. Exp. Biol. Med. (Maywood) 248, 2500 –2513 (2023). https://doi.org/10.1177/15353702231220665 APPENDIX A: Quantum Gate Definitions and Operations This appendix provides a reference overview of the fundamenta...

    work page doi:10.1177/15353702231220665 2023

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.