CoChat

A hybrid deep learning model for detection and mitigation of DDoS attacks in VANETs

Open access
Jayakrishna, Naramalli, Prasanth, N. Narayanan · 2025 · Scientific Reports

The DOI matches its Crossref record; 64 of the 64 references it cites were found in a registry or an index; 14 of the 39 sources its claims cite were checked for support.

Abstract

Intelligent transport systems are increasing in application for real-time communication between vehicles and the infrastructure, and along with that are increasing the popularity of vehicular ad-hoc networks (VANETs). However, the very open and dynamic environment gives rise to varied kinds of DDoS attacks that can disrupt safety-critical services. The existing mechanisms for detection of DDoS attacks in VANETs have been found to suffer from low efficacy of detection, high magnitude of false alarm rates, and poor adaptability to evolving patterns of attacks. To address this challenge, this paper introduces VANET-DDoSNet++, a novel, multi-layered defense framework that uniquely integrates optimized feature selection, advanced deep learning detection, adaptive reinforcement learning mitigation, and secure blockchain-based reporting. The preprocessing step ensures high quality of data by dealing with missing values, removing outliers, augmenting the data, and detecting outliers effectively, preparing for analysis. The features including network traffic statistics, spatiotemporal data, deep traffic embeddings, and behavioural patterns are extracted. To improve the detection performance, a hybrid selection strategy is introduced featuring an adaptive dragonfly algorithm (ADA) and an Enhanced grasshopper optimization algorithm (EGOA) for feature selection where the optimal features are determined. Finally, the detection part applies a hybrid architecture of deep learning referred to as VANET-DDoSNet++, where convolutional LSTM networks, attention layers, and residual/dense connections are used for reliable DDoS detection. An adaptive reinforcement learning-based intrusion mitigation approach with reward shaping tailors defense strategies dynamically with evolving attack vectors by all means. The decentralized trust management mechanism based on blockchain is intended for a secure and verifiable real-time threat reporting from vehicles. The CIC-DDoS2019 dataset, which includes real-world vehicular traffic data with modern reflective DDoS attacks, is utilized for evaluation. The experimental results show that VANET-DDoSNet++ surpasses other currently existing methodologies achieving 98.04% accuracy with 70% training data and 99.18% with 80% training data besides dramatically reducing false positive and negative rates as well as improving overall precision, F1-score, sensitivity, and specificity. The factor deals with the evolution of DDoS attacks whereas VANET networks offer a dynamic and secure intrusion detection and mitigation framework.

Overview

CoChat Summary

TL;DR

The paper proposes VANET-DDoSNet++, a multi-layered framework combining preprocessing, ADA-EGOA feature selection, hybrid deep learning detection, Q-learning reinforcement-learning mitigation, and blockchain-based reporting for DDoS attacks in VANETs. It reports high detection accuracy on CIC-DDoS2019, reaching 98.04% with 70% training data and 99.18% with 80% training data.

Key findings

  • VANET-DDoSNet++ achieved precision and F1-score of 98.70% (70/30 split) and 99.15% (80/20 split), with accuracies of 98.04% and 99.18%; at 80/20 it reported FPR 0.0077, FNR 0.0085, MCC 0.9917, and NPV 99.11%.
  • Hybrid ADA+EGOA feature selection reduced the feature space by 46%, improved F1-score from 0.91 to 0.96 and AUC-ROC from 0.93 to 0.98, selected 81.2% of relevant features, and converged in 48 iterations versus ADA 65 and EGOA 55.
  • The RL-based IPS reported 95% throughput versus rule-based IPS 70%, ADS 80%, and ML-IDS 85%; ~10 ms latency versus ML-IDS ~35 ms; 98% mitigation efficiency versus ML-IDS 91%, ADS 90%, and rule-based IPS 85%; and 3% false-positive rate versus ML-IDS 6% and ADS 8%.
  • Ablation showed the full model at 99.18% accuracy, F1-score 0.9915, and ROC-AUC 0.9983; removing the CNN module reduced accuracy to 95.87%.

Methods

  • Design: a multi-stage pipeline using preprocessing (median/mode imputation, Savitzky-Golay filtering, SMOTE, class-weighted loss, stratified mini-batching, dynamic class reweighting), feature extraction (network traffic, spatiotemporal, deep traffic embeddings via GCN/GAT, behavioral features via BiLSTM), hybrid ADA-EGOA feature selection, ConvLSTM with attention/residual/dense connections for detection, Q-learning RL-IPS for mitigation, and blockchain-based reporting/logging.
  • Data: CIC-DDoS2019 representative subset with more than 50 million DDoS flows and over 80 features; 12 DDoS attack types during training and seven during testing; splits included 70/30 and 80/20, plus 70% training/15% validation/15% testing and 70/10/20 with fivefold cross-validation; SMOTE used with k = 5.
  • Analysis: evaluation used accuracy, precision, F1-score, sensitivity, specificity, FPR, FNR, MCC, NPV, AUC-ROC/AUPR; statistical tests included McNemar’s test (p = 0.0146), Wilcoxon signed-rank test (p = 0.0078), and paired t-tests with p < 0.05; ablation, robustness, cross-validation, and baseline comparisons were also reported.

Limitations

  • The paper states that the multi-layered architecture brings computational overhead and energy expenses, making deployment on resource-limited vehicular devices difficult; it also notes possible increased false positives or false negatives under highly dynamic network conditions, scalability concerns for blockchain consensus at high vehicular density or large scale, and a simulation-to-real-world transferability gap.
  • The paper states that performance is sensitive to dynamic VANET topologies, irregular node density, and fluctuating mobility, and that hybrid or evolving attacks can cause higher false positives; blockchain reporting latency remains a bottleneck, with report validation and block propagation incurring ~2450 ms combined delay, extending beyond 3200 ms under congestion and reducing throughput by up to 25%, even when optimized to ~1800 ms.
  • Evident from the text: the evaluation uses CIC-DDoS2019, which the paper acknowledges is not VANET-specific and lacks GPS coordinates, vehicular mobility traces, and RSU-vehicle interactions; the text also contains inconsistent naming and metrics, such as “Fed-IDMF-VANET” versus “VANET-DDoSNet++,” a 99.4% accuracy claim in the contributions versus 98.04%/99.18% elsewhere, and architecture descriptions mixing CNN+BiGRU with ConvLSTM.

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

Integrity signals

Registry

Crossref · 10.1038/s41598-025-15215-1

Cites

64 of 64 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig.1 Block diagram of the overall proposed methodology.
Fig.1 Block diagram of the overall proposed methodology.CC BY · PubMed Central
Fig. 2 End-to-End AI-Driven VANET security workflow.
Fig. 2 End-to-End AI-Driven VANET security workflow.CC BY · PubMed Central
Fig. 3 Impact of pre-processing on VANET intrusion detection and mitigation.
Fig. 3 Impact of pre-processing on VANET intrusion detection and mitigation.CC BY · PubMed Central
Pseudo-code 1 End-to-end AI-driven VANET security workflow.
Pseudo-code 1 End-to-end AI-driven VANET security workflow.CC BY · PubMed Central
Fig. 4 Architecture of GNN.
Fig. 4 Architecture of GNN.CC BY · PubMed Central
Fig. 5 Impact of feature extraction on the system performance.
Fig. 5 Impact of feature extraction on the system performance.CC BY · PubMed Central
Pseudo-code 2 Hybrid feature selection using ADA and EGOA.
Pseudo-code 2 Hybrid feature selection using ADA and EGOA.CC BY · PubMed Central
Fig.6 Flowchart of the proposed Hybrid Optimization Approach.
Fig.6 Flowchart of the proposed Hybrid Optimization Approach.CC BY · PubMed Central
Fig.7 Impact of optimal feature selection on system performance – AN Explainable AI based analysis.
Fig.7 Impact of optimal feature selection on system performance – AN Explainable AI based analysis.CC BY · PubMed Central
Fig. 8 Convergence analysis.
Fig. 8 Convergence analysis.CC BY · PubMed Central
Fig. 9 Structure of LSTM61.
Fig. 9 Structure of LSTM61.CC BY · PubMed Central
Fig. 10 ConvLSTM architecture.
Fig. 10 ConvLSTM architecture.CC BY · PubMed Central
Fig. 11 Structure of DQN.
Fig. 11 Structure of DQN.CC BY · PubMed Central
Fig. 12 Confusion matrix of 70% and 80% training data.
Fig. 12 Confusion matrix of 70% and 80% training data.CC BY · PubMed Central
Fig. 13 ROC Curve for 70% and 80% training data.
Fig. 13 ROC Curve for 70% and 80% training data.CC BY · PubMed Central
Fig. 14 Precision-recall curve for 70% and 80% training.
Fig. 14 Precision-recall curve for 70% and 80% training.CC BY · PubMed Central
Fig. 15 (a) Recall Vs precision and (b) AUC-ROC.
Fig. 15 (a) Recall Vs precision and (b) AUC-ROC.CC BY · PubMed Central

Verification

Claims the paper makes

5 Valid · 6 Not valid · 11 Not checked · 194 Own results and interpretations

65 more claims · 2 with their cited sources checked

Create a free account to see all 216 claims and to verify the 8 references still open.

See all claims — free account
  • It may state two separate claims as one. 128
  • Probably not a substantial claim. 41
  • The quote only partly states this claim. 1

Create a free account to see these claims.

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.

4 of 4

Cited works that were retracted

Works this paper cites that were later retracted, from the retraction notices publishers deposit. Major when the notice came out before this paper was published.

2 of 2

Citations

Cites

2 retracted

Show all 64 references

Metadata

Authors
Jayakrishna, N., & Prasanth, N. N.
Affiliations
Vellore Institute of Technology University (IN)
Corresponding author
Prasanth, N. Narayanan
Venue
Scientific Reports
Published
2025-10-01
Volume
15
Issue
1
Pages
34170-34170
ISSN
2045-2322 (ISSN-L)
OpenAlex
W4414705919
Cited by
20
References
55
Citations by year
2026: 18 · 2025: 2
FWCI
8.40
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Vellore Institute of Technology, Vellore
Topics
  • Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Vehicular Ad Hoc Networks (VANETs)
  • Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
Keywords
DDoS attack detection, vehicular ad hoc networks, adaptive reinforcement learning, feature selection, dragonfly algorithm, grasshopper optimization algorithm, convolutional LSTM, attention mechanism, decentralized trust management, CICDDoS2019, intelligent transportation systems, false alarm reduction

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

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

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

Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: A hybrid deep learning model for detection and mitigation of DDoS attacks in VANETs. https://app.cochat.ai/papers/10.1038/s41598-025-15215-1
The paper
Jayakrishna, N., & Prasanth, N. N. (2025). A hybrid deep learning model for detection and mitigation of DDoS attacks in VANETs. Scientific Reports. https://doi.org/10.1038/s41598-025-15215-1

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.