Zero-Trust Network Access with Federated Learning for Privacy-Preserving Intrusion Detection in Distributed Communication Systems

Karar Talal (University of Al-Qadisiyah)

Journal of Technology and System Information
Vol. 3 No. 2 (2026): April

The fast-growing distributed communication systems, such as cloud environment, Internet of Things (IoT) platform, and edge computing computers, have greatly compounded the threat of contemporary cybersecurity. The common traditional intrusion detection systems (IDS) are based on centralized collection and analysis of data, which initiates significant issues concerning the privacy of data, scale issues, and communication overheads. This paper will overcome these issues by developing a new Zero-Trust Network Access (ZTNA) model that combines Federated Learning (FL) with privacy-constrained intrusion detection in distributed communication setting. This suggested architecture will be made of three collaborative layers: edge nodes which process local data and train local models, a federated aggregation server which coordinates the global model update using the Federated Averaging (FedAvg) algorithm, and a zero-trust policy engine which dynamically assesses access control decisions based on user trust scores, and network risk assessments. Deep learning techniques, such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and Transformer, are used to create local IDS models with which spatial and temporal patterns of attacks can be effectively detected. The experiments are carried out with well-known datasets of cybersecurity benchmarks that are UNSW-NB15, CSE-CIC-IDS2018 and TON IoT. The environment of implementation makes use of the TensorFlow federated, PyTorch, Docker based edge nodes, and a Kubernetes orchestration framework to recreate realistic distributed conditions. Experimental evaluation proves that the offered framework is much more effective in terms of increasing the accuracy of intrusion detection and decreasing false positive rates and maintaining data privacy. Moreover, federated learning combined with zero-trust policies eliminates centralized dependency of data and improves adaptive control access of network elements in dynamic network context. The findings illustrate how the suggested method has the potential of establishing scalable, privacy conscious, and robust intrusion detection systems in next generation distributed communication networks.