CoChat
Sign Up
Cited by 118Network Security and Intrusion DetectionNetwork Packet Processing and OptimizationAdversarial Robustness in Machine Learning

Security intrusion detection using quantum machine learning techniques

Open access
Kalinin, Maxim, Krundyshev, Vasiliy·2022·Journal of Computer Virology and Hacking Techniques
DOI 10.1007/s11416-022-00435-0PMC PMC9226281
Open this paper in CoChat — free accountOpen PDF

Abstract

Conventional machine learning approaches applied for the security intrusion detection degrades in case of big data input ( $$10^6$$ and more samples in a dataset). Model training and computing by traditional machine learning executed on big data at a common computing environment may produce accurate outputs but take a long time, or produce poor accuracy by quick training, both disparate to malicious activity. The paper observes the quantum machine learning (QML) methods overcoming the barriers of big data and the computing abilities of common hardware for the purpose of high performance intrusion detection. Quantum support vector machine (QSVM) and quantum convolution neural network (QCNN) as concurrent methods are discussed and evaluated comparing to the conventional intrusion detectors running on the traditional computer. The QML-based intrusion detection utilizes our own dataset that implements the grouping of the network packets into the input streams eatable for the QML. We have developed the software solution that encodes the network traffic streams ready to the quantum computing. Experimental results show the ability of the QML-based intrusion detection for processing big data inputs with high accuracy (98%) providing a twice faster speed comparing to the conventional machine learning algorithms utilized for the same task.

Integrity signals

Registry

Crossref · 10.1007/s11416-022-00435-0

Cites

47 of 47 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

11 of 21 cited sources checked · 6 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Quantum computers can provide quadratic or exponential acceleration over conventional ML for large dataset processing.

    “Processing a huge dataset input, the quantum computer offers quadratic and even exponential acceleration against the conventional ML as proposed in [14–16].”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [14] Schuld, M., Sinayskiy, I., Petruccione, F.: An introduction to quantum machine learning. Contemp. Phys. 56(2), 172–185 (2015) source

      Section 3.1 (p. 8) reports a claimed quadratic speed-up over classical algorithms for a specific quantum inner-product method. Section 2.2 (p. 5) frames processing growing amounts of information as a future expectation; the cited passages do not establish exponential acceleration or a general speed-up for huge datasets. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [15] Adcock, J., Allen, E., Day, M., Frick, S., Hinchliff, J., Johnson, M., Morley-Short, S., Pallister, S., Price, A., Stanisic, S.: Advances in quantum machine learning. Science (2015)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [16] Bisio, A., Dariano, G.M., Perinotti, P., Sedlák, M.: Quantum learning algorithms for quantum measurements. Phys. Lett. Sect. A Gen. Atomic Solid State Phys. 375(39), 3425–3434 (2011)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Premisecitation

    Theoretically, training a binary classifier can be exponentially accelerated if the classification algorithm runs on a quantum computer.

    “Thus, theoretically, the training process of a binary classifier can be exponentially accelerated if the described classification algorithm is run on a quantum computer [29].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [29] Nawaz, S.J., Sharma, S.K., Wyne, S., Patwary, M.N., Asaduzzaman, M.: Quantum machine learning for 6g communication networks: state-of-the-art and vision for the future. IEEE Access 7, 46317–46350 (2019) source
  • Findingown data

    QSVM classified network streams with 98% accuracy.

    “Using QSVM, the classification was done with accuracy of 98%.”
    Quote in the paperSubstantial claim
  • Findingown data

    In experiments with traditional SVM on big data, only HTTP Flooding and Port Scanning attacks were detected with high accuracy.

    “When using the conventional SVM, only HTTP Flooding and Port Scanning attacks are detected with high accuracy.”
    Quote in the paperSubstantial claim
  • Findingown data

    For most remaining attack classes, conventional SVM accuracy was between 0.4 and 0.8.

    “The accuracy of most of the remaining classes is in the range from 0.4 to 0.8.”
    Quote in the paperSubstantial claim
  • May combine two claims. 12
  • Classified as probably not a substantial claim. 5

  • May combine two claims.

    Premisecitation

    QSVM is effective for large data, does not overfit, gives high accuracy with big feature maps, and allows use of the kernel trick.

    “Among the above listed methods, QSVM is an effective computing model in case of working with a large amount of data – it does not tend to overfit, provides high accuracy when working with a big feature map, and allows use of the kernel trick [39, 40].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [39] Cortes, C., Vapnik, V.: Support-vector networks. Mach. Learn. 20(3), 273–297 (1995) source

      The paper studies classical support-vector networks, not QSVM. Section 4 describes kernel-based feature-space calculations, and Sections 6.2–6.2.2 report strong accuracy on digit datasets, including one with 60,000 training patterns; Section 6.2.1 says increasing feature-space dimensionality showed no overfitting problems in those experiments. These findings do not establish that QSVM has the stated properties or that it generally does not overfit. Page numbers are unavailable in our copy. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites [40] Havlicek, V., Corcoles, A.D., Temme, K., Harrow, A.W., Kandala, A., Chow, J.M., Gambetta, J.M.: Supervised learning with quantum-enhanced feature spaces. Nature 567(7747), 209–212 (2019) source

      The SVM discussion (p. 11) explains how kernels permit high-dimensional feature maps, and the experimental results (p. 5) report high classification success on small, constructed datasets. Those passages do not establish high accuracy specifically with large feature maps or effectiveness on large datasets. The discussion of VC-dimension bounds (p. 5) also does not establish that QSVM does not tend to overfit. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    QCNN hyperparameters such as the number of convolution and subsampling layers are fixed, and the unitaries themselves are learned.

    “As in a classical CNN, the QCNN hyperparameters such as the number of convolution and subsampling layers are fixed, and the unitaries themselves are learned [31].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [31] Cong, I., Choi, S., Lukin, M.D.: Quantum convolutional neural networks. Nat. Phys. 15(12), 1273–1278 (2019) source

  • May combine two claims.

    Premisecitation

    Existing network intrusion datasets contain packets of two classes, but packet-level classification has drawbacks: many packets are not malicious in content, some attacks spread payload across packets, and staged attacks make packet classification ineffective.

    “However, in practice, the approach of classifying the individual packages has a few significant drawbacks:many packets from the available datasets are not malicious in terms of their content.”
    • Also stated in: § Synthesis of a stream dataset for the quantum classifiers · plain
      “there is a class of attacks that can spread the payload across different packets;many attacks are carried out in stages, and an IDS based on a packet classification is ineffective in this case.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [54] Kang, H., Ahn, D.H., G.M., L., Yoo, J.D., K.H., P., H.K., K.: IoT network intrusion dataset. http://ieee-dataport.org/open-access/iot-network-intrusion-dataset (2019)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [55] Garcia, S., Parmisano, A., Erquiaga, M.J.: IoT-23: A labeled dataset with malicious and benign IoT network traffic (Version 1.0.0). www.stratosphereips.org/datasets-iot23 (2020)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [56] Koroniotis, N., Moustafa, N., Sitnikova, E., Turnbull, B.: Towards the Development of Realistic Botnet Dataset in the Internet of Things for Network Forensic Analytics: Bot-IoT Dataset. (2018). arXiv:1811.00701 source

      Split into 2 atoms — 0 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Prior work has reviewed QSVM for DDoS detection and experimentally studied QSVM on NSL-KDD and NB-15 datasets, showing readiness for intrusion detection.

    “For example, [17] reviews a quantum support vector machine (QSVM) constructed to detect DDoS attacks. [18] presents the experimental study of the QSVM running with conventional NSL-KDD and NB-15 datasets, that denotes the QSVM’s readiness to the intrusion detection on common attack datasets.”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [17] Payares, E., Martinez-Santos, J.C.: Quantum machine learning for intrusion detection of distributed denial of service attacks: a comparative overview. In: Proceedings of SPIE Quantum Computing, Communication, and Simulation, p. 47 (2021) source
    • cites [18] Gouveia, A., Correia, M.: Towards quantum-enhanced machine learning for network intrusion detection. In: 2020 IEEE 19th International Symposium on Network Computing and Applications, pp. 1–8 (2020) source

  • May combine two claims.

    Premisecitation

    QCNN is specified by parameters, corresponds to a double exponential reduction compared to a generic quantum circuit-based classifier, and promises efficient training and implementation.

    “To classify N-qubit input states, a QCNN is thus specified by parameters. This corresponds to a double exponential reduction comparing to a generic quantum circuit-based classifier and promises efficient training and implementation.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [31] Cong, I., Choi, S., Lukin, M.D.: Quantum convolutional neural networks. Nat. Phys. 15(12), 1273–1278 (2019) source

  • Classified as probably not a substantial claim.

    Premisecitation

    QML methods generalize ML methods so that quantum advantages can be used.

    “The QML methods generalize the ML methods in such a way that quantum advantages can be used [28, 29].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [28] Ezhov, A.A., Ventura, D.: Quantum neural networks, pp. 213–235 (2000) source
    • cites [29] Nawaz, S.J., Sharma, S.K., Wyne, S., Patwary, M.N., Asaduzzaman, M.: Quantum machine learning for 6g communication networks: state-of-the-art and vision for the future. IEEE Access 7, 46317–46350 (2019) source

  • May combine two claims.

    Premisecitation

    QCNN has exponential storage capacity, resistance to noise in input data, and high performance.

    “QCNN has exponential storage capacity, resistance to noise in input data and high performance [41].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [41] Kerenidis, I., Landman, J., Prakash, A.: Quantum algorithms for deep convolutional neural networks. (2020). arXiv:1911.01117 source

      The Conclusions (p. 33) report potential speedups and MNIST accuracy similar to a classical CNN, supporting a qualified performance claim. Numerical Simulations (p. 32) show learning despite noise introduced during the algorithm, not resistance to noise in the input data. Quantum Preliminaries (p. 8) describes encoding a vector in a logarithmic number of qubits, but does not establish the claimed exponential storage capacity for the QCNN. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Premisecitation

    Quantum computing technology is one of the most promising tools in the ML area.

    “Instantaneous processing of huge datasets and solving the optimization task make the quantum computing technology one of the most promising tools in the ML area [13].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [13] Zhaokai, L., Xiaomei, L., Nanyang, X., Jiangfeng, D.: Experimental realization of quantum artificial intelligence. (2014). arXiv:1410.1054 source

  • May combine two claims.

    Findingown data

    The authors' stream dataset contains more than 10 million records and 58 fields.

    “As a result, the stream dataset has been developed containing over 10 million records, consisting of 58 fields.”
    • Also stated in: § Conclusion · plain
      “The stream dataset has been developed to improve the QML processing; it contains 58 parameters such as average flag values, minimal, maximal, mean values of packet length and standard deviation, bandwidth bytes/sec, number of packages, flow duration, minimal, maximal, mean interval between bursts and standard deviation, packets per second Layer 4 protocol, stream category.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    QSVM and QCNN can be trained approximately twice as fast as conventional ML on large inputs, and this speed advantage grows with input volume.

    “QSVM and QCNN can be trained approximately twice faster on a big input, and this priority keeps growing if the input data volume enlarges more.”
    • Also stated in: § Conclusion · plain
      “Due to the QML, the training time has been reduced by more than twice.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationnone given

    QCNN can select the most significant features with higher probability, reducing method complexity.

    “The QCNN makes it possible to select the most significant features with a higher probability, which reduces the complexity of the method.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    QCNN is more promising than QSVM despite QSVM being faster.

    “Comparing the QSVM and QCNN detectors, we have concluded that the QCNN is more promising, despite the fact that the QSVM is faster.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    QML-based methods surpassed conventional ML implementations in both accuracy and performance.

    “The QML-based methods have surpassed the ML-based implementations both in accuracy and performance.”
    • Also stated in: § The experimental study · boosted
      “The results of the experiments have demonstrated abilities of the QML against the conventional ML-based detectors when classifying the big volumes of input data.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    Qiskit and Tensorflow Quantum are the most promising platforms and frameworks for implementing quantum computing and QML.

    “Analysis of platforms and frameworks for implementing the quantum computing has shown that the most promising are Qiskit platform and Tensorflow Quantum framework.”
    • Also stated in: § Platforms and software frameworks supporting the QML · plain
      “The most promising kit is Qiskit, which allows us to manage the resources and adapt the developed ML applications for the specific quantum computers.”
    • Also stated in: § Platforms and software frameworks supporting the QML · plain
      “Tensorflow Quantum library was selected for its main advantages: flexibility, ready-to-use ML models, box application packages, scalability in hardware and software, large online community, and compatibility with Keras library.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    QCNN can vary the number of layers and core sizes to solve classification tasks.

    “Another advantage of the QCNN is its ability to vary the number of layers and the size of the cores to solve the classification task.”
    Quote in the paperNot a substantial claim
15 of 15

  • The paper relies on this without support.

    The stream dataset and its encoding preserve the attack-relevant information needed for accurate intrusion detection.

    “To bypass these issues, we introduce the converting a dataset containing attack types and normal traffic into a stream dataset. The fields in the stream are the transformed fields of the packets entering that stream.”
    “To train a QML model on real data, one needs to be able to translate data from a bit representation to a qubit one. For this purpose, the encoder has been developed.”

  • The paper relies on this without support.

    The emulated quantum circuits used in the experiments accurately represent the performance of real quantum hardware.

    “Cirq library executes the quantum circuits;NVidia Cuda provides the faster emulation for quantum circuits;”
    “Tensorflow Quantum contains the framework structures for the QML such as qubits, gates, schematics, and measurement operators.”

  • The paper relies on this without support.

    The labels in the existing and generated network traffic datasets are correct.

    “For large-scale experiments, additional traffic was generated using the nmap and hping programs; packet capture was carried out using the Wireshark program.”

  • The paper relies on this without support.

    The authors' stream dataset is representative of large-scale network environments to which the QML intrusion detection results are generalized.

    “When we have a large-scale network with a big volume of security-relevant data, the QML-based intrusion detection makes the protection more efficient than a traditional ML approach.”
4 of 4

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Conclusion calls QCNN a conventional detector

    Minor

    The conclusion lists QCNN among the conventional detectors, but the experimental section identifies QCNN as the quantum implementation of CNN.

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. Nokia: Nokia catalyzes the next chapter of the internet; innovations deliver massive performance and heightened security needed for cloud and machine era. www.nokia.com/about-us/news/releases/2017/06/14/nokia-catalyzes-the-next-chapter-of-the-internet-innovations-deliver-massive-performance-and-heightened-security-needed-for-cloud-and-machine-era (2021)
  2. Zegzhda, D.P., Stepanova, T.V.: Efficiency estimation of the protection software for neutralization and elimination of botnets. Nonlinear Phenomena Complex Syst. 17(3), 290–295 (2014)
  3. Wiebe, N., Kapoor, A., Svore, K.M.: Quantum Perceptron Models. In: Advances in Neural Information Processing Systems, pp. 4006–4014 (2016)
  4. LeCun, Y., Bengio, Y.: Convolutional Networks for Images, Speech, and Time Series, pp. 276–279 (2003)

Cited by

Sorted by citations

Related

Authors
Kalinin, M., & Krundyshev, V.
Affiliations
Peter the Great St. Petersburg Polytechnic University (RU)
Corresponding author
Kalinin, Maxim
ORCID
Venue
Journal of Computer Virology and Hacking Techniques
Publisher
Springer Science and Business Media LLC
Published
2022-06-24
Volume
19
Issue
1
Pages
125-136
ISSN
  • 1772-9890 (ISSN-L)
  • 2263-8733
DOI
10.1007/s11416-022-00435-0
PMC
PMC9226281
OpenAlex
W4283373641
Semantic Scholar
327e0fe8cf65aafb440f159a3095b6c2d7a446b8
Cited by
118
References
47
Citations by year
2026: 43 · 2025: 31 · 2024: 28 · 2023: 14 · 2022: 2
FWCI
8.32
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Computer Virology and Hacking Techniques (publishedVersion, other-oa)
  • PubMed Central (publishedVersion)
Funders
Ministry of Science and Higher Education of the Russian Federation (075-15-2021-1333, 075-15-, Priority 2030")
Topics
  • Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
  • Physical Sciences › Computer Science › Hardware and Architecture › Network Packet Processing and Optimization
  • Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
Keywords
intrusion detection, quantum machine learning, quantum support vector machine, quantum convolutional neural networks, big data, malicious activity detection, quantum computing

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-09-28.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-09-28.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review