Unleashing the Power of Diffusion Models for Imbalanced Data in Intrusion Detection Systems

Federated Learning (FL) has emerged as a promising solution to train IDS models in a distributed manner while preserving data privacy. However, the heterogeneous nature of client data often leads to imbalanced training datasets, causing degraded model performance. This work addresses the challenge of handling non-iid data in FL for Intrusion Detection Systems (IDS). To overcome this, we propose the integration of a Diffusion Model (DM), a generative AI, to generate synthetic data locally on each client device. This approach augments the existing non-iid data with diverse, high-quality samples to improve model generalization and convergence. Our findings highlight the potential of using generative AI for data augmentation in federated settings and open new avenues for improving the robustness of distributed intrusion detection systems.