AD-5GIoT: AI-Based Anomaly Detection System for 5G-IoT Networks

In the context of a 5G-IoT communication environment, devices and users interact through the Internet, exposing them to numerous anomalies arising from diverse cybersecurity challenges including DDoS attacks, malware, and man-in-the-middle attacks. Consequently, it is essential to develop robust anomaly detection (AD) solutions specifically designed for 5G-IoT applications to safeguard them against these cyber threats. In this study, we propose an AI-based AD system that utilizes real-time traffic data for 5G-IoT networks, designed specifically for critical sectors such as healthcare, smart grid, industry 4.0, etc. The proposed AD system comprises three synchronized components aimed at enhancing the cybersecurity of 5G-enabled systems. These components include the 5G-IoT communication infrastructure, an AI-enabled AD engine, and an anomaly dashboard. To evaluate the performance of the proposed system, we conducted experiments using two AI models: Graph Neural Network (GNN) and Convolutional Neural Network (CNN). These experiments were performed on real-time 5G data, which included both benign and malicious traffic, generated over a 5G wireless network. The 5G-IoT anomaly detection system was evaluated using feature subsets, for <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$k=10$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$k=25$</tex>. The results showed that with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$k=10$</tex> and using the GNN learning model, the overall accuracy achieved was 99.19%. For the benign case, the precision was 98.86%, while for the malicious case, the precision was even higher at 99.71%. From our analysis, it can be concluded that the proposed system using GNN demonstrates promising results for binary classification in real-time 5G -IoT anomaly detection.