Intrusion Detection in IoT Networks Using Federated Learning
Neivaldo I. Matos Filho, Alex Vitorino, Muhammad Ahsan, Fernanda Sumika Hojo de Souza, Daniel L. Guidoni · 2025
This work investigates the use of Federated Learning (FL) techniques for intrusion detection in IoT device networks, analyzing the behavior of the FedAvg, FedProx, and FedAdam algorithms under different partitioning and classification strategies. The experiments employed the CICIoT2023 dataset, which includes realistic attack scenarios with significant class imbalance. Preprocessing steps, training with a 1D-CNN model, and comparative performance analysis were conducted using metrics such as accuracy, F1-score, and execution time. The results show that FedProx exhibits greater stability in the presence of data heterogeneity, while FedAvg is sensitive to non-uniform distributions. FedAdam, in turn, remained stable in simpler tasks but showed inconsistent performance in more complex scenarios. The analysis highlights that the choice of aggregation function directly impacts model robustness in distributed environments, emphasizing the importance of appropriate strategies to ensure performance and generalization in FL-based intrusion detection systems.