A Federated Learning and Deep Learning-Based Intrusion Detection Mechanism for IoT Networks
Kapil Shrivastava, Manish Tiwari, Prasun Chakrabati · 2025
The rising number of Internet of Things (IoT) devices made scalable and privacy preserving network security a significant concern. This study introduces a Federated Learning (FL) based Intrusion Detection System (IDS) employing deep learning for enhanced performance, integrating both centralized and federated learning to protect IoT environments. We have utilized Long Short-Term Memory (LSTM) model and 1 Bi-directional LSTM (BiLSTM) which enables the proposed system to efficiently capture temporal and contextual links in data sourced from the CIC-IDS2017 dataset. Preprocessing steps including data cleaning and normalization are applied to ensure high model efficiency and accuracy. In the centralized setup, BiLSTM achieved the best performance with 97.31 % accuracy, 96.98% precision and 96.73% recall. The federated BiLSTM model demonstrated 96.79% accuracy, 96.50% precision and 96.17% recall, showing minimal trade-off while preserving user privacy. Averaged over 100 epochs the centralized BiLSTM maintained 94.97% accuracy, while the federated counterpart achieved 93.89%, indicating the resilience of federated learning even under non-IID data settings. These results validate that federated learning can deliver high-performing and privacy-preserving IDS for IoT networks with performance comparable to centralized models. The framework demonstrates strong potential for real-world deployment where sensitive data cannot be centrally aggregated.