A Federated GAN-Augmented Explainable Deep Learning Framework for IoT Intrusion Detection

The increasing number of Internet of Things (IoT) devices has broadened the area vulnerable to attack, making networks more susceptible to advanced cyber threats. A major obstacle to creating strong Intrusion Detection Systems (IDS) is the significant imbalance in real-world network data. In this data, normal traffic greatly exceeds attack instances, which leads to poor detection rates for less common attacks. This paper suggests a new approach that combines Federated Learning (FL), Generative Adversarial Networks (GANs), and Explainable AI (XAI) to overcome these challenges. A Wasserstein GAN with Gradient Penalty (WGAN-GP), improved using Genetic Algorithms, is used to create realistic synthetic attack data in a privacy-focused, collaborative manner across various IoT devices. This enhanced data is then used to train a combined deep learning model, using both Convolutional and Long Short-Term Memory networks (CNN-LSTM), to identify spatial and temporal patterns in network activity. Additionally, Local Interpretable Model-agnostic Explanations (LIME) are used to provide clear, understandable explanations for each prediction. When tested on several standard datasets (UNSWNB15, IoT-23, CIC-IDS2018), the proposed approach shows a significant improvement in performance, achieving over 99% accuracy and better detection of uncommon attacks, while ensuring data privacy and clear explanations for security analysts.