Generative Adversarial Networks for Improving Detection of Network Intrusions in IoT Environments
Saurabh Pahune, Puneet Matapurkar, Sumeet Mathur, Himanshu Sinha
Abstract
The Internet of Things (IoT) connects everyday devices, enabling data exchange with minimal human intervention. Nevertheless, strong Intrusion Detection Systems are required since IoT networks are very susceptible to cyber assaults. This study examines an employ of ML and DL techniques for IDS using the CICIDS2017 dataset. Data preprocessing involved normalization, one-hot encoding, and SMOTE to address class imbalance, alongside feature selection to enhance model efficiency. In IoT networks, the generator and discriminator are the two primary parts of the suggested Generative Adversarial Networks (GANs) for IDS. The GAN-based approach outperforms traditional ML and DL models, such as SVM and Decision Trees, by offering an accuracy of 99.4%, precision of 99.25%, recall of 99.6%, and F1 score of 99.4%. These outcomes highlight the potential of GANs in securing IoT systems by accurately detecting threats while optimizing computational demands, paving the way for efficient, real-time IoT security applications.