Enhanced Machine Learning Ensemble Approach for Securing Small Unmanned Aerial Vehicles From GPS Spoofing Attacks

Abstract

Unmanned aerial vehicles (UAVs) substantially rely on the utilization of global positioning systems (GPS) to navigate. A simulator for commercial GPS applications with false GPS signals can lead to the deviation of a GPS-guided drone from its planned path. As a result, an anti-spoofing technology is required to assure UAV operating safety. Several approaches have been developed to detect GPS spoofing, however, predominantly such methods rely on additional hardware. Using additional hardware might not be an ideal solution for small and low-capacity UAVs. Detecting signal spoofing attacks in small UAVS has significant importance. This study presents a stacked ensemble approach to detect GPS signal spoofing within the context of small UAVs. The initial phase involves outlining the sequential procedures for obtaining and preparing the GPS signal dataset, including details about the UAV hardware, blocker, data collection timing, environmental factors, and the utilization of z-score normalization for preprocessing. Then controlled simulation tests with varying experimental conditions are conducted and the model is built using a support vector machine and convolutional neural network. Additionally, a comprehensive comparative assessment is conducted to analyze the efficacy of the proposed model against traditional machine learning models. Experimental results demonstrate notably good performance by the proposed model with a 99.74% accuracy, showing its superior performance in the context of GPS signal spoofing in small UAVs.