Enhancing GPS Accuracy with Machine Learning: A Comparative Analysis of Algorithms

Abstract

In the realm of wireless communications, the Global Positioning System (GPS), integral toGlobal Navigation Satellite Systems (GNSS), finds extensive applications ranging fromvehicle navigation to military operations, aircraft tracking, and Geographic InformationSystems (GIS). The reliability of GPS is often compromised by errors particularly prevalentin dense and structurally complex environments, where signal attenuation by environmentalobstacles like mountains and buildings is common. These challenges necessitate thedeployment of high-cost, precision GPS receivers capable of enhanced signal tracking andacquisition. This study investigates the reduction of GPS positioning errors by implementinga machine learning framework, utilizing a dataset from vehicle tracking devices equippedwith Novatel and Ublox technologies. Ten machine learning prediction algorithms wereevaluated, focusing on techniques that introduce randomness for stability, employ proximityfor predictions, incorporate regularization to prevent overfitting, and leverage both singleand ensemble methods to refine analyses. Among the evaluated algorithms, the Extra Treesalgorithm was distinguished by its superior performance, achieving a coefficient ofdetermination (R²) of 99.6%, with the lowest error rates compared to its counterparts. Theerrors were quantified as Root Mean Square Error (RMSE) at 1.01E-4, Mean Absolute Error(MAE) at 4.14E-5, and Mean Square Error (MSE) at 1.03E+0 for normalized data. Acomparative assessment across ten scenarios demonstrated that the machine learning-enhanced approach deviated by approximately 6.8 meters on average, markedly improvingaccuracy over traditional GPS methods and reducing positional deviations to a scale ofmeters. This advance represents a significant stride towards minimizing GPS inaccuraciesin complex environments, providing a robust framework for enhancing navigationalprecision in critical applications.