How Quantum Is the Advantage? A Fair, Calibration- and Noise-Aware Benchmark and Attribution Audit of Quantum Machine Learning for Network Intrusion Detection
Quantum machine learning (QML) for network intrusion detection (NIDS) is routinely reported to reach near-perfect accuracy, yet the most rigorous studies find that well-tuned classical models remain competitive, and that apparent quantum gains may be artefacts of classical dimensionality reduction and implicit regularisation rather than genuine quantum effects. We ask not whether a quantum model can post a high accuracy, but how quantum the advantage really is. We present a unified, reproducible QML-IDS benchmark evaluating hybrid variational quantum circuits and quantum-kernel SVMs against five honestly-tuned classical baselines across four standard NIDS datasets (NSL-KDD, UNSW-NB15, CICIDS2017, NF-ToN-IoT-v2) under one leakage-controlled protocol, with an equal-budget feature view, imbalance- and calibration-aware metrics with significance testing, and a simulated NISQ noise sweep. We introduce a quantum-attribution audit (parameter-matched classical controls, a random-feature kernel, and a regularisation sweep) that quantifies how much of any gain is genuinely attributable to the quantum component. Tuned classical models (Random Forest, XGBoost) match or exceed the quantum models on aggregate detection on every dataset, and the audit attributes this to classical preprocessing and regularisation rather than quantum effects. Two advantages survive false-discovery-rate correction: the quantum-kernel SVM out-ranks its direct classical surrogate (a random-feature kernel) on AUPRC and ROC-AUC, and a small four-qubit hybrid out-detects the best classical baseline at the 1% false-positive operating point on the distribution-shifted NSL-KDD task (p = 0.005, BH q = 0.030). Code, seeds, and splits are released; our contribution stands whether quantum wins, ties, or loses.
Quantum machine learning (QML) for network intrusion detection (NIDS) is routinely reported to reach near-perfect accuracy, yet the most rigorous studies find that well-tuned classical models remain competitive, and that apparent quantum gains may be artefacts of classical dimensionality reduction and implicit regularisation rather than genuine quantum effects. We ask not whether a quantum model can post a high accuracy, but how quantum the advantage really is. We present a unified, reproducible QML-IDS benchmark evaluating hybrid variational quantum circuits and quantum-kernel SVMs against five honestly-tuned classical baselines across four standard NIDS datasets (NSL-KDD, UNSW-NB15, CICIDS2017, NF-ToN-IoT-v2) under one leakage-controlled protocol, with an equal-budget feature view, imbalance- and calibration-aware metrics with significance testing, and a simulated NISQ noise sweep. We introduce a quantum-attribution audit (parameter-matched classical controls, a random-feature kernel, and a regularisation sweep) that quantifies how much of any gain is genuinely attributable to the quantum component. Tuned classical models (Random Forest, XGBoost) match or exceed the quantum models on aggregate detection on every dataset, and the audit attributes this to classical preprocessing and regularisation rather than quantum effects. Two advantages survive false-discovery-rate correction: the quantum-kernel SVM out-ranks its direct classical surrogate (a random-feature kernel) on AUPRC and ROC-AUC, and a small four-qubit hybrid out-detects the best classical baseline at the 1% false-positive operating point on the distribution-shifted NSL-KDD task (p = 0.005, BH q = 0.030). Code, seeds, and splits are released; our contribution stands whether quantum wins, ties, or loses.
Quantum machine learning (QML) is increasingly investigated for network intrusion detection, but the practical value of current quantum models remains uncertain because many studies rely on small simulated data sets, idealized execution, weak classical baselines, or isolated accuracy comparisons. This study develops and evaluates a resource-aware hybrid quantum-classical intrusion detection framework that combines fixed fidelity quantum kernels, repaired Quantum Kernel Alignment (QKA), a Variational Quantum Classifier (VQC), LightGBM, prototype-based kernel reduction, and probabilistic stacking. Experiments use the UNSW-NB15 benchmark under a leakage-aware preprocessing protocol and repeated evaluation across five random seeds. The experimental program covers 4- and 6-qubit representations, multiple training sizes, prototype budgets, and finite-shot configurations, yielding 260 experimental configurations/runs. The strongest quantum-assisted model, a 6-qubit QKA-VQC-LightGBM stack at 100 samples per class, obtains mean F1=0.8480, MCC=0.7092, ROC-AUC=0.9398, and PR-AUC=0.9302. Repaired QKA improves F1 from 0.8043 for the corresponding fixed kernel to 0.8089, while 50-prototype QKA attains F1=0.8257 with a more favorable kernel-workload trade-off than larger prototype budgets. Finite-shot experiments between 1,024 and 4,096 shots remain stable in simulation. However, full-data LightGBM remains stronger overall (F1=0.8957, MCC=0.8068, ROC-AUC=0.9755, PR-AUC=0.9704), and corrected statistical tests do not establish general quantum superiority. These findings motivate a Selective Quantum Intrusion Detection System (S-QIDS), in which classical inference handles high-confidence traffic and quantum processing is reserved for uncertain, novel, or high-risk observations. The study therefore reframes near-term quantum intrusion detection from unconditional quantum advantage toward measurable quantum utility, quantum-classical complementarity, and resource-aware quantum intervention.
Intrusion Detection Systems (IDSs) are a cornerstone of modern cybersecurity; however, their performance is increasingly constrained by high-dimensional network traffic and the scalability limitations of classical machine learning approaches. Quantum Machine Learning (QML) has recently emerged as a promising paradigm by leveraging quantum feature spaces and hybrid quantum–classical optimization. In this work, we present a unified and reproducible benchmarking framework for QML-based intrusion detection under realistic Noisy Intermediate-Scale Quantum (NISQ) constraints. The proposed framework integrates hybrid Quantum Neural Networks (QNNs) and Quantum Support Vector Machines (QSVMs) within a consistent experimental pipeline, enabling a systematic comparison of quantum models, embedding strategies, and dimensionality reduction techniques. Experiments are conducted on the NSL-KDD benchmark dataset for both binary (normal versus attack) and multiclass intrusion detection tasks. Multiple quantum embedding strategies, including angle, amplitude, and Instantaneous Quantum Polynomial (IQP) embeddings, are evaluated in conjunction with PCA-based dimensionality reduction to analyze trade-offs between expressivity, circuit complexity, and computational cost. The proposed models are benchmarked against strong classical baselines under identical preprocessing and evaluation conditions. The results demonstrate that QNN- and QSVM-based IDS models achieve competitive and, in several scenarios, superior performance compared to classical approaches, particularly in multiclass settings characterized by nonlinear and imbalanced data distributions. In addition, we provide an in-depth analysis of scalability limitations, training complexity, and NISQ feasibility, highlighting practical considerations for real-world deployment. To ensure full reproducibility, all implementation details, datasets, and experimental configurations are publicly available at: https://github.com/raidanis/QML-IDS-NSLKDD. These findings position hybrid QML as a promising complementary paradigm for next-generation intrusion detection systems, particularly in low-dimensional and high-complexity feature spaces.
Intrusion detection systems (IDS) are essential for identifying cyber threats in complex digital environments. Machine learning (ML) is widely used to improve IDS by detecting anomalies, but classical ML methods often struggle with high-dimensional data and evolving threats. Quantum machine learning (QML) has been proposed as a potential paradigm to overcome some of these limitations, but is constrained by noisy intermediate-scale quantum (NISQ) challenges, affecting quality. This study systematically evaluates three QML models, Pegasos Quantum Support Vector Classifier (QSVC), Variational Quantum Classifier (VQC), and a Hybrid Quantum–Classical Neural Network (HQNN), for network anomaly detection. The models were optimized and tested on the ToN_IoT and NSL-KDD datasets using IBM quantum simulators under both ideal and noisy conditions. Performance was analyzed through the F1-score distribution as a function of circuit complexity, revealing how entanglement and noise affect robustness across different backends. Comparisons with classical models contextualize the current maturity of QML for cybersecurity, while computational time was used as an indicator of model complexity to explore accuracy–efficiency trade-offs. Among all configurations, Pegasos-QSVC achieved the best results, with 94.60% accuracy and an F1-score of 94.13%. The findings provide practical guidelines for designing noise-resilient QML models and highlight their potential for reliable intrusion detection under realistic quantum conditions.
Variational quantum circuits are said to buy richer decision boundaries for fewer parameters than a classical network needs, and intrusion detection on network flows is among the tasks where that claim is made most often. What backs it up is thinner than the publication count suggests: hybrid quantum IDS papers typically show one training run against loosely specified baselines, leaving no way to separate an architectural effect from run-to-run variance. We report a controlled evaluation of a hybrid quantum- classical detector, Q-NIDS, across three configurations sharing one preprocessing pipeline. Two use CICIDS2017 with a six-qubit circuit combining data re-uploading, a learnable entanglement topology and quantum recurrent memory gates; the third scales to eight qubits on BigFlow-NIDS, a 66.9-million-flow NetFlow corpus, using a batched state-vector simulator matching PennyLane's lightning.qubit to 9.7 × 10-8. On CICIDS2017 the calibrated model reaches 87.97% accuracy and 49.09% macro F1, exceeding a 1D-CNN (40.52 ± 0.83%), an LSTM (39.06 ± 2.04%) and an MLP (37.98 ± 3.97%) trained on identical data, but falling 12.47 points short of a Random Forest (61.56 ± 0.20%). Three findings cut against the usual story. Most of the margin over the neural baselines is bought by post-hoc prior adjustment, a classical correction worth 8.11 macro-F1 points. The learnable entanglement topology never trains: a non-differentiable branch in gate selection cuts its gradient path, after which an unopposed entropy regulariser drives the adjacency matrix to a value where every conditional gate switches off. On BigFlow-NIDS an ablated control with the quantum layer deleted reaches 10.79% macro F1 against the full model's 12.53%, a margin too narrow to credit to quantum processing. We report all of it, failed configurations included, and argue that controls of this kind belong in any hybrid quantum machine learning claim.
Network intrusion detection systems (NIDS) must classify high-dimensional, heavily imbalanced traffic flows under tight operational constraints, and the flow features produced by modern capture tools routinely exceed the qubit budgets available on today’s noisy intermediate-scale quantum (NISQ) hardware. Existing quantum machine-learning (QML) detectors typically either assume that every extracted feature can be encoded on a dedicated qubit or apply generic dimensionality reduction that is blind to the downstream quantum encoder; in both cases the qubit budget — the binding constraint on real devices — never drives which features are encoded, and evaluation is frequently confined to ideal statevector simulation that omits shot noise, depolarizing error, and readout confusion. This paper presents ANHQ-IDS, an adaptive, noise aware hybrid quantum intrusion-detection framework that makes the qubit budget a first-class design constraint. A resource-constrained feature-selection stage scores candidate features through a composite objective jointly balancing mutual-information relevance, inter-feature redundancy, cross-split stability, and an explicit quantum-encoding cost, and greedily selects a subset sized to a target qubit budget rather than fixed a priori. The retained features are embedded through parameterized angle and second-order (ZZ) feature maps and classified by a hardware-efficient variational quantum classifier (VQC) trained with gradient-free optimizers (COBYLA and SPSA) under finite-shot sampling. To characterize behaviour under realistic hardware conditions, the classifier is evaluated on a statevector reference and under synthetic depolarizing and readout-error models, and benchmarked on CIC-IDS2017 against classical baselines (logistic regression, random forest, support vector machine) using a leakage-resistant train/validation/test protocol. On a class-balanced CIC-IDS2017 benchmark, the ZZ-encoded VQC attains macro-F1 0.766 and ROC-AUC 0.853 under exact simulation — exceeding logistic regression (0.502) and approaching a support-vector baseline, while a random forest remains strongest (0.964); performance degrades gracefully under depolarizing and readout noise, consistent with the predicted (1−p) contraction. All preprocessing, feature-selection, and quantum-training components are released as a reproducible Qiskit package with fixed seeds and recorded circuit-resource metadata (qubits, depth, gate and parameter counts).
Under tight feature and compute budgets, classical threat detection pipelines often degrade on near-decision-boundary events. Small quantum processors are now available, but existing work inadequately shows whether quantum components improve end-to-end threat detection under the above resource constrained conditions. This paper tries to address this gap with a hybrid architecture that uses a compact multilayer perceptron layer to compress the information in data and then routes the processed features to a few qubit quantum heads implemented in quantum support vector machine (QSVM) and variational quantum circuit (VQC) models. On a simulation platform, we benchmark these hybrid models against classical models with comparable parameter budgets on two representative cybersecurity tasks, network intrusion detection on NSL-KDD dataset and spam filtering on Ling-Spam dataset. To validate their precision on real quantum hardware, we deploy the best 4-qubit QSVM model on an IBM Quantum device with noise-aware execution, evaluated on a smaller sub-dataset. In the results, shallow quantum heads consistently match, and on difficult near-boundary cases modestly reduce missed attacks and false alarms compared to classical models using the same features. Hardware validation results track the simulation behavior closely enough that the remaining gap is dominated by device noise rather than model design. Furthermore, we conduct adversarial attacks to test the robustness of one QSVM model. Taken together, the study shows that even on small, noisy devices, carefully engineered quantum components may function as competitive, budget-aware components in practical cyber threat detection applications.