<![CDATA[The increasing reliance on Global Positioning System (GPS) technology in Unmanned Aerial Vehicles (UAVs) has exposed them to cybersecurity threats, particularly GPS spoofing attacks that manipulate location data. This study explores the effectiveness of various machine learning-based approaches in detecting GPS spoofing in UAV communication networks. Supervised classification models, unsupervised anomaly detection techniques, and deep learning-based autoencoders are evaluated to determine their capability in identifying spoofed signals. The dataset used for training and testing contains multi-dimensional UAV network parameters with labeled GPS spoofing instances. Experimental results indicate that traditional anomaly detection models, such as Isolation Forest, One-Class SVM, and Local Outlier Factor, struggle with detection accuracy and exhibit high false-positive rates. The autoencoder-based approach achieves the highest accuracy (91.20%) but has poor precision (3.97%) and recall (4.73%), highlighting limitations in threshold selection and anomaly classification. Computational complexity analysis reveals that deep learning models, despite their accuracy advantages, require significant computational resources, making them less feasible for real-time UAV applications. This study identifies critical challenges in GPS spoofing detection, including dataset bias, environmental variability, and model hyperparameter sensitivity.]]>
Copyrights © 2025
