CoChat
Sign Up
Cited by 364Privacy-Preserving Technologies in DataNetwork Security and Intrusion DetectionInternet of Things and AI

The DOI matches its Crossref record; 176 of the 177 references it cites were found in a registry or an index; none of the 50 sources its claims cite is checked for support yet.

Federated Deep Learning for Cyber Security in the Internet of Things: Concepts, Applications, and Experimental Analysis

Open access
Ferrag, Mohamed Amine, Friha, Othmane, Maglaras, Leandros, Janicke, Helge, Shu, Lei·2021·IEEE Access
DOI 10.1109/access.2021.3118642
Open this paper in CoChat — free accountOpen PDF

Abstract

In this article, we present a comprehensive study with an experimental analysis of federated deep learning approaches for cyber security in the Internet of Things (IoT) applications. Specifically, we first provide a review of the federated learning-based security and privacy systems for several types of IoT applications, including, Industrial IoT, Edge Computing, Internet of Drones, Internet of Healthcare Things, Internet of Vehicles, etc. Second, the use of federated learning with blockchain and malware/intrusion detection systems for IoT applications is discussed. Then, we review the vulnerabilities in federated learning-based security and privacy systems. Finally, we provide an experimental analysis of federated deep learning with three deep learning approaches, namely, Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), and Deep Neural Network (DNN). For each deep learning model, we study the performance of centralized and federated learning under three new real IoT traffic datasets, namely, the Bot-IoT dataset, the MQTTset dataset, and the TON_IoT dataset. The goal of this article is to provide important information on federated deep learning approaches with emerging technologies for cyber security. In addition, it demonstrates that federated deep learning approaches outperform the classic/centralized versions of machine learning (non-federated learning) in assuring the privacy of IoT device data and provide the higher accuracy in detecting attacks.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

This paper surveys federated deep learning for IoT cyber security and adds an experimental comparison of centralized vs. federated DNN, CNN, and RNN intrusion detection models on three IoT traffic datasets. It concludes that federated learning improves privacy and can approach or outperform centralized learning, though its own reported centralized accuracies are often higher than the federated global-model results.

Key findings

  • The review covers federated learning-based security and privacy systems across Industrial IoT, edge computing, Internet of Drones, Internet of Healthcare Things, Internet of Vehicles, cloud computing, 5G-enabled IoT, mobile crowdsensing, and cyber-physical systems; it also discusses federated learning with blockchain and federated intrusion/malware detection.
  • In centralized experiments, the highest accuracies were RNN on Bot-IoT at 96.76%, DNN on MQTTset at 90.06%, and RNN on TON_IoT at 99.98%; the lowest were DNN on Bot-IoT at 95.76%, RNN on MQTTset at 89.29%, and CNN on TON_IoT at 98.87%.
  • In federated experiments, IID global models approximated centralized performance; Non-IID models struggled initially but after 50 rounds approached centralized performance. For Bot-IoT with CNN and K=15, the worst client had 01.00% accuracy at the first round, 52.98% at round 50, while the global model achieved 90.35%.
  • The paper states that federated deep learning approaches outperform classic/centralized machine learning in assuring IoT data privacy and provide higher accuracy in detecting attacks, although the experimental comparison text reports that federated global models “managed to approach” centralized performance.

Methods

  • Design: comprehensive survey plus experimental analysis training three deep learning IDS models—DNN, CNN, and RNN—under centralized and federated learning settings.
  • Data: three real IoT traffic datasets—Bot-IoT (69.3 GB PCAP, over 72 million records; CSV files used), MQTTset, and TON_IoT—with preprocessing including dropping selected features, one-hot encoding of `proto`, and Z-score normalization.
  • Federated setup and analysis: Sherpa.AI framework on Google Colaboratory with NumPy, Pandas, TensorFlow, and Keras; client distributions K=5, K=10, and K=15; IID and Non-IID data distributions; 50 federated rounds; metrics include accuracy, precision, recall, F1-score, and ROC curves.

Limitations

  • Paper-stated open challenges include: privacy can still be revealed by analyzing the global model; differential privacy can degrade performance or require high computing power; IoT network constraints such as bandwidth, interference, noise, and limited reliability; data bias and unequal data sizes across devices; and lack of FL-specific IoT frameworks.
  • Evident from the text: the experiments use pre-existing public datasets with engineered partitions to mimic federations because of a lack of FL-specific datasets; experiments were run on Google Colab; source code is available only upon request; and the conclusion that federated learning provides higher attack-detection accuracy is not fully supported by the paper’s own centralized-vs-federated accuracy results, which show federated global models approaching rather than consistently exceeding centralized models.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/access.2021.3118642

Cites

176 of 177 references verified0 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 of 50 cited sources checked · 0 supported · 0 flagged

Show all 42 claims

  • It may state two separate claims as one. 20
  • Probably not a substantial claim. 8
26 of 26

  • The paper relies on this without support.

    Many cyber security researchers have difficulty in finding the best learning type (centralized or federated) to test and evaluate their proposed security methods in IoT applications.

    “Therefore, many cyber security researchers have difficulty in finding the best learning type (i.e., centralized or federated learning) to test and evaluate their proposed security methods in IoT applications, and selecting an appropriate federated deep learning method is an essential issue in this field.”

  • The paper relies on this without support.

    Selecting an appropriate federated deep learning method is an essential issue in this field.

    “Therefore, many cyber security researchers have difficulty in finding the best learning type (i.e., centralized or federated learning) to test and evaluate their proposed security methods in IoT applications, and selecting an appropriate federated deep learning method is an essential issue in this field.”

  • The paper relies on this without support.

    Traditional centralized learning has issues such as privacy, latency, bandwidth, and connectivity.

    “However, there are some issues for traditional centralized learning such as privacy, latency, bandwidth, and connectivity.”

  • The paper relies on this without support.

    Many IoT devices are susceptible to attack through unsafe design, deployment, and configurations.

    “Many of those devices, however, are susceptible to attack through unsafe design, deployment, and configurations.”

  • The paper relies on this without support.

    Federated learning can suffer from machine learning attacks against the locally created models.

    “Since it is trained on sensitive user data, federated learning can suffer from machine learning attacks against the locally created models.”

  • The paper relies on this without support.

    FL-based tasks require the data distribution to be Non-IID and unbalanced, reflecting the properties of the real-world scenario.

    “FL-based tasks require the data distribution to be Non-Independent and Identically Distributed (Non-IID) and unbalanced, which reflects the properties of the real-world scenario.”

  • The paper relies on this without support.

    Adversaries can use cyber security attacks such as adversarial attacks or poisoning attacks to degrade the accuracy of a machine learning model or deduce if an IoT device has been involved in some mission from their local model updates.

    “These adversaries can use cyber security attacks such as adversarial attacks or poisoning attacks to degrading the accuracy of a machine learning model or deduce if an IoT device has been involved in some mission from their local model updates.”
7 of 7

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

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.

  • FedAGRU is attributed to different cited authors

    Minor

    One passage attributes the proposed FedAGRU system to Cvitic et al. [29] [88], while an earlier passage says Cetin et al. [96] proposed FedAGRU.

  • Dataset citations do not match the numbered references

    Minor

    The dataset discussion cites [149] for TON_IoT, but reference 149 is titled "LEAF: A benchmark for federated settings"; the TON_IoT reference is numbered 150. The same one-number mismatch affects the cited Bot-IoT and MQTTset references.

  • TON_IoT figure lists six classes after stating five

    Minor

    The discussion says Fig 16 presents ROC curves for five TON_IoT classes, but the figure caption enumerates classes 0 through 5, including XSS as a sixth class.

  • Research questions are attributed to the learning-performance figure

    Minor

    The text points to Fig 17 for the research questions, but the caption identifies Fig 17 as learning performances.

  • MQTTset figure lists six classes after stating five

    Minor

    The discussion says Fig 15 presents ROC curves for five MQTTset classes, but the figure caption enumerates classes 0 through 5, including Malformed as a sixth class.

5 of 5

Citations

Cites

Flagged first, then by citations

Research works printed but not linked: 5

  1. McMahan, 2017, Communication-efficient learning of deep networks from decentralized data, Artificial Intelligence and Statistics, 1273
  2. 2021, Federated Learning: Collaborative Machine Learning Without Centralized Training Data
  3. Schneble, Attack detection using federated learning in medical cyber-physical systems, 1
  4. Goodfellow, Generative adversarial nets, 27, 1
  5. 2021, The MNIST Database of Handwritten Digits

Cited by

Sorted by citations

See all 364

Related

See all

Metadata

Authors
Ferrag, M. A., Friha, O., Maglaras, L., Janicke, H., & Shu, L.
Affiliations
  • Edith Cowan University (AU)
  • Badji Mokhtar-Annaba University (DZ)
  • De Montfort University (GB)
  • Nanjing Agricultural University (CN)
Corresponding author
Ferrag, Mohamed Amine
ORCID
Venue
IEEE Access
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2021
Volume
9
Pages
138509-138542
ISSN
2169-3536 (ISSN-L)
DOI
10.1109/access.2021.3118642
OpenAlex
W3203932036
Semantic Scholar
82f305d3cb1a82fb6d4a5fcabb4fda40cb2b05e0
Cited by
342
References
174
Citations by year
2026: 53 · 2025: 94 · 2024: 80 · 2023: 70 · 2022: 44 · 2021: 1
FWCI
17.16
Type
journal article
Language
en
Open access
Licence
Locations
  • IEEE Access (publishedVersion, cc-by)
  • DMU Open Research Archive (De Montfort University) (submittedVersion, cc-by)
  • Australasian Journal of Paramedicine (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Edinburgh Napier Research Repository (Edinburgh Napier University) (submittedVersion, other-oa)
  • Greater South Information System
Funders
Cyber Security Cooperative Research Centre (CSCRC) and Edith Cowan University, Australia
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
  • Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
  • Physical Sciences › Computer Science › Information Systems › Internet of Things and AI
Keywords
federated deep learning, Internet of Things security, IoT privacy, malware detection, intrusion detection, blockchain, Industrial Internet of Things, edge computing, Internet of Drones, Internet of Medical Things, Internet of Vehicles, Bot-IoT dataset, ToN-IoT, recurrent neural networks, convolutional neural networks, deep neural network

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.

A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

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