Hybrid VAE-GAN framework for synthetic IoT traffic generation and intrusion detection
|
Full Text |
Pdf
|
|
Author |
Alakunta Vyshnavi and D. Mahendra Reddy
|
|
e-ISSN |
1819-6608 |
|
On Pages
|
796-809
|
|
Volume No. |
21
|
|
Issue No. |
12
|
|
Issue Date |
August 20, 2026
|
|
DOI |
https://doi.org/10.59018/062685
|
|
Keywords |
internet of things security, synthetic data generation, VAEGAN, network simulation, intrusion detection system (IDS), multi-class classification, generative AI, flow-level traffic modeling.
|
Abstract
The IoT networks have continued to face severe issues in intrusion detection that are linked to the absence of data, class imbalances and the inadequacies of simulator-generated traces. The paper introduces SynthNet, which is a hybrid framework that VAE-GAN-based data generating framework, which is assisted by the simulators of IoT to enhance the performance of multi-class intrusion detection. The generative module is being trained to generate latent representation of flow-level IoT traffic and the generated normal and attack samples are used to augment the training distribution. Such simulated flows are added along with traffic provided by the simulator to form an augmented data to train a multi-class IDS. The suggested solution is demonstrated to be efficient because, according to the experimental analysis, SynthNet has the accuracy of 0.998848, the macro F1-score of 0.9649, and the weighted F1-score of 0.9989 in a 7-class IoT dataset. The confusions matrices, ROC and precision recall curve show that the class separability is near perfect and probability distribution of representative test flow is tracing of well-calibrated predictions with the correct one among the classes being dominant (Normal Traffic 0.9918 probability). These results indicate that the hybrid generative model and IoT simulation are quite effective to improve the generalization of the entire classifier, rare-attack detection, and the diversity of the data, which is a feasible and comprehensive solution to the IoT security analytics.
Back