Federated Learning-Based Intrusion Detection System for Enhancing Internet of Things Security

Abstract

The widespread deployment of Internet of Things (IoT) devices has increased the surface area for cyberattacks, making effective intrusion detection a growing necessity. While Machine Learning (ML) and Deep Learning (DL) techniques have improved Intrusion Detection Systems (IDS), traditional centralized approaches often compromise data privacy by requiring raw data transmission to a central server. Federated Learning (FL) offers a privacy-preserving alternative by enabling decentralized model training without exposing raw data. In this work, we implement a federated stacking ensemble Intrusion Detection System (FedStack-IDS), combining an Artificial Neural Network (ANN) and Random Forest (RF) as base learners, with a Logistic Regression (LR) model as the metaclassifier. The approach is evaluated using the CICIoT2023 dataset in both centralized and decentralized settings. FedStack-IDS achieved an accuracy of 87.45% and a macro F1-score of 88%, outperforming existing FL-based IDS solutions, while also maintaining strong detection on rare attack classes. Moreover, communication overhead was reduced by 75% through quantization, confirming the system’s effectiveness for accurate, efficient, and privacy-preserving intrusion detection in IoT environments.