The Internet of Things (IoT) has dramatically expanded global connectivity, with billions of heterogeneous devices generating massive volumes of network traffic. This growth has significantly enlarged the attack surface for cyber threats, particularly synthetic unknown attacks that exploit vulnerabilities not yet captured in existing signature databases. Traditional intrusion detection systems (IDS), which rely on known attack signatures, fail to detect such novel threats. This study proposes a comprehensive Adversarial Machine Learning (AML) framework designed to detect synthetic unknown attacks in IoT networks through the integration of three complementary techniques: Generative Adversarial Networks (GANs) for synthesizing realistic attack samples, Adversarial Autoencoders (AAEs) for robust low-dimensional feature extraction, and Deep Neural Networks (DNNs) for real-time anomaly classification. The framework was evaluated on a controlled synthetic IoT dataset comprising 250,000 traffic samples. Results demonstrate detection accuracy of 94.8%, precision of 93.1%, recall of 92.7%, and F1-score of 92.9%, substantially outperforming signature-based IDS (63.5%), SVM (87.2%), and Random Forest (90.1%). The framework maintained detection rates of 90.2%, 87.5%, and 85.3% under FGSM, PGD, and Carlini-Wagner adversarial perturbations, respectively. Scalability testing across datasets up to 4 million samples confirmed sustained performance with inference latency below 0.021 seconds per sample. The proposed framework represents a scalable, adversarially resilient approach to IoT intrusion detection, with limitations regarding real-world generalization openly acknowledged.
The rapid adoption of Internet of Things (IoT)-enabled applications has significantly expanded the cyberattack surface across a wide range of critical systems such as industrial IoT (IIoT), smart grids, transportation, healthcare, industrial control systems, and smart cities. False data injection attack (FDIA) has emerged as a serious security threat to these applications due to its stealthiness and adversarial nature, silently corrupting the data integrity of critical operational processes without triggering conventional detection mechanisms. Existing FDIA solutions rely on single-model architectures that are built based on classical or limited predefined attack scenarios. Such solutions often fail to achieve robust detection under adversarial and evolving attack conditions; accordingly, they lack generalisability and are insufficient to capture the broader scope of FDIAs. In this study, a hybrid detection framework is proposed that integrates a Random Forest classifier with an unsupervised anomaly detection model based on a deep autoencoder combined through a Logistic Regression metaclassifier. The proposed framework addresses the gap in single-model detectors that either rely on fixed decision boundaries that struggle with gradually evolving stealthy FDIA patterns or on anomaly detection that lacks strong discriminative power in separating subtle adversarial deviations from normal operational variability. Different types of stealthy and adversarial FDIA have been modelled and injected into the dataset samples for use in training the proposed model. The results show that the overall detection performance of the proposed architecture improved by 2.39 percentage points in terms of F1-score while maintaining a low false-positive rate of 0.49%. These findings reflect the effectiveness of feature representation learning via autoencoders and hybrid classification strategies against stealthy and adversarial FDIA patterns. Future work should include temporal modelling for further advancing robust detection against evolving adversarial threats.
A. Albarrak, Fuad A. Ghaleb, Sultan Noman Qasem et al.· Italian National Conference...· 0 citations
The rapid expansion of internet of things (IoT) networks has heightened security risks, particularly regarding distributed denial of service (DDoS) attacks against devices with limited computing capacity. High detection accuracy is crucial for these resource-constrained environments, where false positives can disrupt legitimate traffic and false negatives allow attacks to persist. However, modern reinforcement learning (RL) and machine learning (ML) intrusion detection solutions often exhibit poor generalization due to static state representations. To address this, this paper proposes a deep Q-learning (DQL) framework that integrates K-means clustering directly into the RL action space. Unlike prior RL-based IDS models, our approach dynamically integrates clustering into the learning process, enabling adaptive state representation and improved generalization to unseen traffic patterns. The system is formulated as a Markov decision process where the agent optimizes a composite reward function based on accuracy, precision, recall, and F1-score. Evaluated on the N-BaIoT dataset using 10-fold cross-validation, the proposed method achieves a classification accuracy of 98.95% and a weighted F1-score of 98.73%, significantly outperforming traditional ML and RL baselines. These results demonstrate the framework's effectiveness as a scalable, adaptive solution for intelligent IoT DDoS detection.
Lana Kamla Ahmed, Kayhan Zrar Ghafoor· Indonesian Journal of Electr...· 0 citations
A constraint-aware adversarially robust Internet of Things (IoT) traffic classification system with protocol validity, device behavior consistency, and manifold-aware training and evaluation is presented in this study. In realistic IoT communication semantics, resilience as a constrained min–max optimization problem allows adversarial perturbations. Comprehensive testing on sample IoT traffic datasets shows that baseline models achieve 95.1% accuracy under benign conditions but plummet following hostile attacks. The proposed defense reduces untargeted attack success rates to <18% while achieving 81.3% accuracy at ε=0.05 and 70.6% at ε=0.10. The proposed constraint-aware adversarial framework significantly enhances IoT traffic classification by achieving 97.4% accuracy and maintaining 90.6% robustness at ε=0.10, outperforming state-of-the-art methods. It reduces attack success rates to 11.2% (untargeted) and 7.9% (targeted) through protocol-compliant perturbations and manifold-aware learning. Additionally, the model achieves an efficient trade-off with 21.4 ms latency and 650 flows/sec throughput, making it suitable for real-time edge deployment. These results demonstrate improved robustness, realism, and deployability compared to existing approaches.
Intrusion Detection Systems (IDSs) are essential for monitoring network traffic and identifying malicious activities in modern cyber-physical, Internet of Things (IoT), enterprise, and distributed network environments. However, developing reliable IDS models remains challenging because attack behaviors evolve over time, realistic datasets are difficult to obtain, traffic records may be incomplete, attack classes are often imbalanced, and privacy constraints limit centralized data collection. Recent advances in generative artificial intelligence (AI) and Federated Learning (FL) provide new opportunities to address these limitations. Generative models can support anomaly detection, synthetic traffic generation, data augmentation, data imputation, adversarial traffic generation, and IDS alert explanation. FL enables distributed IDS training without directly sharing local network traffic, making it suitable for privacy-sensitive and geographically distributed environments. This survey provides a structured review of generative AI and FL techniques for IDS. We first summarize representative IDS research directions, including adversarial machine learning, anomaly-based detection, IoT-oriented IDS, explainable IDS, and benchmark datasets. We then categorize generative AI applications in IDS according to model families and task objectives, covering autoencoder-based models, Generative Adversarial Networks (GANs), diffusion models, and Large Language Models (LLMs). Finally, we review emerging studies that integrate generative AI with FL-based IDS and discuss open challenges, including synthetic data quality, realistic traffic generation, dual-use adversarial risks, non-IID client distributions, communication-efficient model sharing, federated IDS benchmarking, and domain-specific LLMs for network security.
Jiefei Liu, A. S. Tayeen, Pratyay Kumar et al.· 0 citations
The expansion of Internet of Things (IoT) deployments has widened the attack surface available to adversaries, and the man-in-the-middle (MITM) attack remains one of the most damaging threats facing these networks. In a MITM attack, two parties that believe they are communicating directly are in fact exchanging traffic through an intermediary that silently alters or observes the exchange. Common realizations of this threat include ARP spoofing, DNS hijacking, and SSL stripping, each producing a distinct signature in network traffic that a classifier can learn to recognize. This study evaluates and compares five ensemble learning algorithms, Random Forest, Extra Trees, XGBoost, CatBoost, and LightGBM, for the detection of MITM activity in the TON_IoT network traffic dataset. Performance is assessed using accuracy, precision, recall, F1-score, area under the ROC curve, and computational cost. CatBoost obtained the highest detection accuracy (99.2%) and F1-score (0.987), while LightGBM required roughly one third of CatBoost’s training time at a negligible cost in detection quality. Across all five algorithms, boosting methods showed a small but consistent advantage over bagging methods, and detection was effective for every MITM technique considered, with SSL stripping proving the most difficult to identify. The results suggest that the choice among these algorithms in an operational deployment should depend on whether the priority is raw detection accuracy, inference speed, or interpretability, rather than on accuracy alone.
Yasser AbdelSatar, Fatma El-Zahraa Mohamed, Shimaa AbdelNasser et al.· Engineering Systems and Inte...· 0 citations
Network Intrusion Detection Systems (NIDS) play a vital role in modern cybersecurity by leveraging artificial intelligence (AI) in particular deep learning (DL) and machine learning (ML) to detect and mitigate malicious activities. However, these AI-driven systems are highly vulnerable to adversarial attacks, where small, imperceptible perturbations in input data can deceive models and significantly reduce detection accuracy. This raises critical concerns about the security and reliability of intrusion detection, especially in real-world scenarios where attackers exploit adversarial transferability to bypass defenses. This research investigates the threat posed by black-box adversarial attacks via surrogate models, focusing on the ability of adversarial examples to transfer across different architectures. This study simulates real-world adversarial threats, demonstrating how attacks crafted on one model can effectively deceive another, compromising NIDS security. A comparative study is conducted on two widely used AI models: an Artificial Neural Network (ANN) and a Convolutional Neural Network (CNN), both trained on the CICIDS 2019 dataset. The study evaluates the robustness of these models against two gradient-based adversarial attack methods, Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD), to determine their susceptibility under black box adversarial conditions. Experimental results indicate that CNN-based NIDS are more vulnerable to adversarial attacks than ANN-based models, with adversarial examples successfully transferring across architectures. These findings highlight the critical risks associated with adversarial transferability, underscoring the need for enhanced security measures to strengthen AI-driven intrusion detection systems against evolving cyber threats.
Aasim Zafar, Shazra Wali, Sheikh Burhan Ul Haque· International Journal of Inf...· 0 citations