DDoS Attack Detection using Machine Learning for UAV Networks

Attacks targeting network components and devices pose a significant risk to the continuity and performance of network services. Consequently, innovative technical solutions are required to mitigate these threats and enhance the efficiency of attack detection systems. This paper introduces a Distributed Denial of Service (DDoS) attack detection algorithm specifically designed for mobile Unmanned Aerial Vehicles (UAV) networks. Leveraging a Software-Defined Networks (SDN) architecture, we propose a Machine Learning (ML) based detection module that employs a novel set of features, including flow count, speed of source IP, speed of flow entry, source IP address entropy, and destination IP address entropy, to capture both static and movement-driven traffic variations effectively. Furthermore, we derive the correlation between these features at various access points (APs). This helps establish that the attack traffic pattern remains consistent as attacker nodes move between APs. Our detection module employs three ML classifiers, Random Forest (RF), Multi-Layer Perceptron (MLP), and Support Vector Machine (SVM), and our performance evaluation under diverse traffic types (TCP, UDP, and ICMP) and attack intensities (light and heavy) demonstrates that the MLP model outperforms the other two, achieving an average accuracy of around 99%, compared to approximately 95% for state-of-the-art schemes. These results underscore the robustness of our detection framework in both static and dynamic UAV network scenarios, paving the way for developing efficient, resource-aware attack mitigation systems.