GNN Approach for Detecting Network Anomalies in Wireless Systems

Abstract

In today’s interconnected environments, maintaining the security and integrity of wireless networks is more essential than ever. Conventional anomaly detection methods frequently prove inadequate in the context of rapidly changing network architectures, particularly with the emergence of IoT and diverse devices. This project presents an innovative system for wireless network anomaly detection that utilizes Graph Neural Networks (GNNs), especially a hybrid framework that merges Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs). By representing network entities as graph nodes and their interactions as edges, the system proficiently learns both local and global structural patterns to detect anomalies in real time. The execution includes a dynamic web dashboard created with Streamlit, which displays detection results, anomaly distribution, and network health metrics like accuracy and false positives. Achieving over 94% detection accuracy and incorporating a modular pipeline that facilitates live training, data upload, and result examination, this project reveals the potential of GNN-based models to enhance the defense mechanisms of wireless networks.