A Machine Learning Based Smartphone App for GPS Spoofing Detection

Abstract

With affordable open-source software-defined radio (SDR) devices, the security of civilian Global Position System (GPS) is at risk of spoofing attacks. Spoofed GPS signals from SDR devices have indicated that spoofed signals have higher values of signal-to-noise ratios (SNRs). Utilizing these values along with other parameters, we propose a machine learning (ML) based GPS spoofing detection system for classifying spoofed signals. To build our detection system, we launch spoofing attacks on a GPS receiver using a low-cost SDR device, LimeSDR, and apply ML algorithms on SNR values and the number of tracked and viewed satellites. A performance comparison between different ML algorithms shows that Random Forest (RF) and Support Vector Machine (SVM) achieve 99.5% accuracy, followed by K-Nearest Neighbors (KNN) (99.4%). To demonstrate easy integration of the algorithm with GPS enabled devices, we develop an Android-based smartphone app that successfully notifies the user about the spoofing signals.