IoTTraffic: Assessing Generative Models for Internet of Things Attack Data Flows

Analyzing the data generated by gadgets is crucial for spotting and minimizing cyberattacks on the Internet of Things. However, public data representing real attacks still tends to be scarce. To increase data availability, this work presents a study on the use of Generative Adversarial Networks (GANs) to generate synthetic attack data on IoT devices with high fidelity to real data, i.e., with similar characteristics. Simultaneously, ensuring privacy and that the utility of synthetic data in machine learning tasks is similar to real data. For this purpose, two GAN models, CTGAN and NetShare, were compared using a dataset containing normal traffic and attacks on IoT devices. The results indicate that both GAN models are efficient in generating synthetic data, both in fidelity and quality. However, CTGAN proves to be the most efficient model, considering execution time and memory consumption.