Intrusion Detection Using Federated Learning in Non-IID Data Environments

Intrusion Detection Systems (IDS) are essential for protecting networks by detecting threats. Machine learning (ML) based approaches have enhanced the effectiveness for IDS. However, it raises privacy concerns for IDS based on ML approaches due to the need for centralized data. To address this issue, Federated Learning (FL) has been applied to IDS. It allows ML models to be trained on decentralized devices/clients without sharing raw data. However, FL faces challenges with non-independent and identically distributed (non-IID) data, which reduces the performance of intrusion detection models. This paper introduces a loss function that jointly learns compact local representations on each client and a global model across all clients, enhancing the robustness of FL in non-IID data environments. Experimental results demonstrate that our method significantly improves the accuracy and robustness of FL systems for IDS in non-IID environments.