Federated Learning-Based Intrusion Detection Model for Privacy-Preserving Cybersecurity in IoT Networks
Abstract
The growing number of Internet of Things (IoT) devices grants unauthorized people the potential to breach smart systems for more control, leading to major privacy concerns. Centralized Intrusion Detection Systems (IDS) collect and analyze network traffic data, which poses a privacy risk because sensitive data can be leaked. This paper proposes a privacy-preserving Intrusion Detection Model based on Federated Learning (FL) for IoT networks. The model gives the ability to the distributed IoT nodes to train collaboratively, a deep learning-based IDS, which ensures the confidentiality of the data and the absence of raw data sharing. The model combines an IoT Edge Gateway-based lightweight Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) model. This approach was evaluated using the CICIDS and UNSW-NB15 datasets and was subjected to a heterogeneous distribution of data samples. The federated model achieved detection rates of 97.8% and 96.9% for precision, 97.3% for recall, and 97.1% for the F1-score, while for the other models, data exposure was reduced by $100\%$. When compared to systems that adopt a centralized learning approach, the proposed FL-IDS achieved a privacy risk reduction of 35% by model parameter aggregation optimization and maintained the Accuracy of other models to an identifiable extent of 1.2%. The findings indicate that federated learning is an efficient and growing method for applying intelligent and secure solutions for IoT technological environments. The proposed model combines the distributed intelligent and training mechanism models of privacy and secure intelligent computing for next-generation smart systems. The model proposed detects and defends intrusions and maintains the sovereignty of user data; thus, it can be applied to smart city, smart healthcare, and smart industrial IoT environments.