SMARTY: A Real-Time Graph Neural Network Approach for Malicious Traffic Detection
The increasing complexity of cyberattacks makes traditional approaches ineffective for the timely detection of malicious network traffic. Graph Neural Networks (GNNs), while showing strong potential in traffic analysis, still face limitations related to scalability, latency and applicability in real-time scenarios.
The SMARTY project addresses these challenges by introducing a new graph construction methodology designed to efficiently represent relationships between network flows and enable fast, accurate threat detection. The goal is to make GNNs usable in high-performance contexts while maintaining high accuracy and low latency.
Experimental results highlight the model’s ability to identify different types of attacks already in the early stages of communication, even with reduced observation windows and show superior performance compared to a state-of-the-art approach in real-time detection scenarios.
This work represents a step forward toward the integration of AI into next-generation network infrastructures and opens the way to deployment on Data Processing Units (DPUs) for validation in operational environments. The contribution, developed by CNIT and SSSA, was accepted at IEEE ICMLCN 2025, confirming the scientific relevance of the results.
Follow SMARTY for updates:
LinkedIn: https://www.linkedin.com/company/smartyeu/?viewAsMember=true
Website: https://www.smarty-project.eu