<p>With the rapid advancement of the Internet of Things (IoT), ensuring the security of interconnected devices has become a critical challenge. Traditional intrusion detection systems (IDS) often struggle to meet IoT requirements due to limited computational resources, high communication overhead, and centralized data processing. To address these limitations, this study introduces a federated learning–based intrusion detection framework that enables collaborative model training across distributed IoT devices while preserving data privacy. In the proposed system, local models are trained directly on device data, and only model parameters are shared for global aggregation, ensuring both efficiency and confidentiality. The framework is designed to handle both IID and non-IID data distributions, effectively mitigating the heterogeneity inherent in IoT networks. Three deep learning architectures–Convolutional Neural Network (CNN), Deep Neural Network (DNN), and Transformer–are implemented under both centralized and federated settings. Using the Edge-IIoTset dataset, comprehensive experiments evaluate the detection performance and robustness against a wide range of real-world cyberattacks. Results demonstrate that the proposed federated approach achieves superior intrusion detection accuracy, resilience, and privacy preservation compared to conventional centralized models, underscoring its potential for robust and scalable IoT security.</p>