FEDGAN-IDS: Privacy-preserving IDS using GAN and Federated Learning

Abstract

Federated Learning (FL) is a promising distributed training model that aims to minimize the data sharing to enhance privacy and performance. FL requires sufficient and diverse training data to build efficient models. Lack of data balance as seen in rare classes affects the model accuracy. Generative Adversarial Networks (GAN) are remarkable in data augmentation to balance the available training data. In this article, we propose a novel Federated Deep Learning (DL) Intrusion Detection System (IDS) using GAN, named FEDGAN-IDS, to detect cyber threats in smart Internet of Things (IoT) systems; smarthomes, smart e-healthcare systems and smart cities. We distribute the GAN network over IoT devices to act as a classifier and train using augmented local data. We compare the convergence and accuracy of our model with other federated intrusion detection models. Extensive experiments with multiple datasets demonstrates the effectiveness of the proposed FEDGAN-IDS. The model performs better and converges earlier than the state-of-the-art standalone IDS.