Improving IoT Intrusion Detection Using GAN-Based Synthetic Traffic Augmentation

In the field of intrusion detection systems, nonuniform class distribution in training datasets compromises the machine learning based detection mechanisms. Generative Adversarial Networks (GANs) offer promising solutions to address this problem. This paper employs a tailored Conditional Tabular GAN (CTGAN) to generate new data points aligned with the minority-class distribution. The experiments utilize the NF-v2-Bot-IoT dataset for performance evaluation. The results show that employing this tailored architecture to generate synthetic samples for the minority class of theft significantly increases the detection rate from 42% to impressive 99%. This highlights the efficacy of the tailored architecture in substantially enhancing the detection performance for skewed classes.