ENHANCING UAV SECURITY: A HYBRID MACHINE LEARNING APPROACH TO INTRUSION DETECTION IN UAV NETWORKS

Abstract

The widespread integration of Unmanned Aerial Vehicle (UAV) in intelligent transportation, military reconnaissance, and critical infrastructure monitoring has introduced heightened vulnerabilities to sophisticated cyber-physical attacks. Conventional Intrusion Detection System (IDS) typically isolate cyber and physical layers, failing to capture correlations between telemetry anomalies and network threats, and thus struggle to detect multi-vector attacks spanning both domains. This study proposes a hybrid deep learning-based IDS that fuses synchronized telemetry and MAVLink traffic to model spatiotemporal intrusion signatures with high fidelity. The architecture integrates one-dimensional Convolutional Neural Network (1D-CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Temporal Convolutional Network (TCN), jointly capturing local anomalies, sequential dependencies, and long-range deviations, with the final intrusion decision obtained via probability-level ensemble averaging across the three branches. A cyber-physical dataset was developed using ArduPilot Softwarein- the-Loop (SITL) across one-, two-, and four-UAV swarm deployments under three topologies: drone-to-base station (D2BS), leader-follower, and drone-to-drone (D2D) relay. Four representative attack types were included-Denial of Service (DoS), Replay, False Data Injection (FDI), and Evil Twin. The dataset will be made publicly available upon publication to support reproducibility and further research. Beyond cyber-physical fusion, this work addresses key swarm-specific challenges absent in prior studies: syniii chronizing asynchronous data streams and accounting for response-time feasibility in multi-hop UAV communications. Experimental results show over 95% classification accuracy with reduced false positives, outperforming state-of-the-art unimodal baselines. These contributions highlight not only the role of cyber-physical integration but also the importance of swarm-scale evaluation and open datasets in advancing resilient, real-time UAV security systems.