Combining Generative Adversarial Networks (GANs) With Gaussian Noise for Anomaly Detection in Internet of Things (IoT) Traffic

Abstract

This study presents an innovative approach for anomaly detection in Internet of Things (IoT) network traffic based on Generative Adversarial Networks (GANs). To evaluate the model's performance, the CICIDS2017 dataset, which includes various attack types and normal network traffic, was used. The preprocessing process involved feature scaling, the addition of Gaussian noise to enhance model generalization, and the extraction of the Hurst self-similarity parameter to analyze the dynamic behavior of the data. The proposed model consists of a generator that produces pseudo-real data and a discriminator capable of distinguishing between real and fake data. This structure enables the identification of anomaly patterns in IoT traffic data. Performance evaluation demonstrated that the proposed method achieved an accuracy of 99.88%, a recall of 99.88%, in anomaly detection, significantly outperforming traditional detection methods. The main innovation of this research lies in the combination of GAN with the calculation of the Hurst parameter and the addition of noise to the input data, improving the model's ability to detect complex attacks, including low-frequency and zero-day attacks. The results indicate that this model offers superior performance in learning attack patterns, enhancing detection accuracy, and reducing false positives. This approach can serve as a powerful tool in Intrusion Detection Systems for the security of IoT networks.