Hybrid Graph Neural Network for Scalable Network Intrusion Detection
Vijayalakshmi T
Network intrusion detection systems (NIDS) face scalability challenges with growing network traffic and complex attack patterns. This paper proposes a hybrid graph neural network (HGNN) that combines graph convolutional networks (GCN) with recurrent layers for efficient feature extraction and temporal modeling. By representing network flows as dynamic graphs, HGNN captures spatial dependencies and sequential behaviors, enabling real-time anomaly detection on large-scale datasets. Experiments on benchmark datasets like CIC-IDS2018 demonstrate superior accuracy (97.8% F1-score) and 5x faster inference compared to traditional deep learning baselines, making it ideal for high-volume environments.