Abstract:
In the connected digital world of today, sophisticated cyberattacks pose a growing threat to distributed networks. The Core networks and Edge both are a challenging task in provision of smooth and seamless services, whereas the utilization of Information Technology (IT) is at its highest. The span of Networks is increasing day by day and accordingly security threats to service providers, organizations, Telecommunications even all IT setups till national and international level are also increasing. With this, the Network Security is a huge milestone to achieve. Detecting sophisticated threats is a challenge for traditional centralized Network Intrusion Detection Systems (NIDS), particularly in distributed situations where real-time anomaly detection is essential. This research proposes a Distributed Hybrid deep learning-based NIDS using anomaly detection framework AI-HGF-IDS that combines Generative Adversarial Networks (GANs) with Deep Neural Networks (DNN) and Long-Short-Term-Memory (LSTM) for threat detection and Federated Learning (FL) for distrusting the detected anomaly across the network. This model increases the overall detection efficiency of network threats making it an exceptional Distributed NIDS. The proposed system leverages GANs for generating synthetic data to simulate advanced attack scenarios, LSTM for sequential anomaly detection, and DNNs for extracting deep patterns from network traffic. Federated Learning enables nodes in a distributed network to collaborate by sharing only anomaly detection updates, thereby preserving privacy and scalability. Using the NSL-KDD dataset for validation, the AI-HGF-IDS model demonstrates its capability to detect known and unknown anomalies in real time, offering a scalable, privacy-preserving, and highly effective solution for securing distributed networks. The proposed framework addresses the limitations of conventional NIDS by offering a decentralized approach that adapts to the dynamic nature of modern networks, ensuring timely and accurate detection of emerging threats. The proposed model achieved 93.45% accuracy as state-of-the-art hybrid approach.
