An Evidence-based Approach for Gps Accuracy Classification

This paper investigates the accuracy of a GPS device. The GPS accuracy is treated as a pattern recognition problem. Each location estimate is classified into a certain accuracy class. Various observation conditions provided by the GPS device are used as features relating a location estimate to an accuracy band. In this paper we introduce an evidence-based classifier (EBC) in which three independent classifiers are used: namely, feed forward neural network, K-nearest neighbor and the support vector machine. The decisions of these classifiers are combined by a reasoning-based-engine using dempster-shafer (DS) evidence theory for decision fusion. The DS engine will produce the final classification decision. As proof of concept, a comprehensive experimental work including two use-cases is conducted in this paper. Experimental results are discussed at the end of this paper.