Comparative Study of Quantum and Classical Machine Learning Models in Binary Classification
Anand Kumar Mishra, Ramanuj Awasthi
quant-ph · cs.LG
Quick verdict
VQC/QSVM vs three classical baselines on Breast Cancer data: classical wins; honest negative result, but a single 45-sample test split limits conclusions.
Summary
This paper presents a controlled empirical comparison of two quantum classifiers — a Variational Quantum Classifier (VQC) and a Quantum Kernel Support Vector Machine (QSVM) — against Logistic Regression, RBF-kernel SVM, and a small MLP on the Breast Cancer Wisconsin (Diagnostic) dataset. All five models consume the same 4-component PCA projection of 30 features, min-max scaled to [0, π]; experiments use a stratified 150-sample subsample (70/30 split, 45 test cases with 17 malignant) simulated in PennyLane on classical hardware. Classical Logistic Regression achieves the best accuracy (97.8%), classical SVM and QSVM tie at 95.6% with identical precision/recall/F1, the classical MLP reaches 93.3%, and the VQC trails at 88.9%. Notably, the authors candidly flag that Table 1's recall treats benign as positive: recomputed per class, Logistic Regression, SVM, and QSVM each identify 16/17 malignant cases (94.1%) while the VQC identifies only 12/17 (70.6%), the worst of the five. Training costs under simulation are high for quantum models (QSVM 23.29 s, VQC 6.46 s vs ≤0.004 s for linear baselines). The main limitations are a single tiny test split with no cross-validation, confidence intervals, or significance testing; no documented hyperparameter tuning; and a VQC budget (30 epochs, MSE loss) that leaves underfitting unruled out. Conclusions are therefore suggestive rather than statistically supported.
Review complete
Assessment at a glance
4
risk flags identified
● high 1
● medium 2
● low 1
4 strengths · 4 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- On a PCA-4-reduced, 150-sample subsample of the Breast Cancer Wisconsin dataset, classical Logistic Regression achieves the best accuracy (97.8%), ahead of classical SVM and QSVM (95.6% each) and the VQC (88.9%).
- Quantum models are far more expensive to train under classical simulation, with QSVM at 23.29 s and VQC at 6.46 s versus ≤0.004 s for the classical linear models.
- The VQC's 100% recall applies to the benign class only; its malignant-class recall is the worst of all five models at 70.6% (12/17), while Logistic Regression, SVM, and QSVM each reach 94.1% (16/17).
- Near-term QML classifiers have not surpassed well-tuned classical counterparts on small structured datasets, consistent with prior comparative studies [3][5].
Strengths
Honest class-conditional reanalysis
The authors themselves flag that Table 1's recall treats benign as the positive class (§5.3) and recompute malignant-class recall from the confusion matrices (§5.4), reporting their own VQC as the worst model (70.6%, 12/17) rather than the flattering 100% benign recall.
Controlled comparison design
All five models consume the same PCA-4, [0, π]-scaled input on identical stratified splits (§3.2-3.3); parameter counts are documented (VQC 16 vs MLP 49, §3.3-3.4), and classical-quantum results are directly comparable.
Public code release
Transparent cost accounting
Weaknesses
Single tiny test split without uncertainty
Tuning claimed but not documented
VQC training budget may drive the headline result
Unremarked QSVM/RBF-SVM equivalence
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper (arXiv:2609.23476) asks whether two near-term quantum classifiers — a 4-qubit Variational Quantum Classifier (VQC) and a Quantum Kernel SVM (QSVM) — can match three classical baselines (Logistic Regression, RBF-kernel SVM, an 8-neuron MLP) on binary classification of the Breast Cancer Wisconsin (Diagnostic) dataset. The central claims, stated in the abstract and Sections 5-6, are: (1) classical Logistic Regression wins on accuracy (97.8% vs 95.6% for classical SVM and QSVM; 88.9% for VQC); (2) quantum models are far costlier to train under classical simulation (QSVM 23.29 s, VQC 6.46 s vs ≤0.004 s for linear models); (3) the VQC's seemingly perfect recall is a benign-class artifact and its malignant-class recall (70.6%, 12/17) is the worst of all five models; (4) therefore near-term QML shows no advantage on small structured datasets. Claim (4) is a negative result stated with appropriate hedging, and the paper is commendably upfront about the metric framing behind claim (3).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 4 strengths & 4 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
Did any quantum machine learning model beat the classical models in this paper?
No. Classical Logistic Regression achieved the best accuracy (97.8%), classical SVM and the quantum kernel SVM (QSVM) tied at 95.6%, and the variational quantum classifier (VQC) was lowest at 88.9% on a 45-sample test set.
Why does the paper report 100% recall for the quantum VQC if the VQC performed worst?
The 100% recall is for the benign class only, because Table 1 treats benign as the positive class. When recall is recomputed for malignant tumors — the clinically critical direction — the VQC detected just 12 of 17 (70.6%), the worst of all five models, while Logistic Regression, SVM, and QSVM each detected 16 of 17 (94.1%).
Was real quantum hardware used in this study?
No. All quantum circuits (4 qubits) were built in PennyLane and simulated classically on Google Colab; testing on real hardware such as IBM Quantum is listed as future work.
Is the code and data for this study available?
Yes. The dataset is the standard Breast Cancer Wisconsin set built into scikit-learn, and the authors state that the implementation, figures, and results are released in a public GitHub repository.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1-7 appear only as captions in the extracted text; the loss curve (Fig. 7) and confusion matrices (Fig. 6) could not be independently inspected, though Table 1's numbers are internally consistent with the reported counts.
- The abstract in the source metadata is truncated mid-sentence; this review relies on the full body text, which contains the complete statements.
- The cited GitHub repository could not be inspected; its contents (e.g., whether the split seed is pinned) are assumed as described in §3.6.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.