Analysing Probabilistic Support Vector Machine Based Localization In Wireless Sensor Networks

Localizing sensors in a sensor network is one of the severe bottlenecks that must be dealt with, before exploiting these kinds of networks efficiently. While there has been many techniques and methods proposed for the issue, most of them suffer from low accuracy, or impose extra costs to the network. A Support Vector Machine (SVM) based method has already been proposed which uses machine learning techniques to achieve a fairly accurate estimate of the location of the nodes. In this paper, we propose to use probabilistic SVM, which is more powerful than the existing method. Moreover, an innovative post processing step called ARPoFiL will be proposed that provides even more improvement to the accuracy of the location of the sensor nodes. We will show analytically and experimentally that probabilistic SVM integrated with ARPoFiL completely outperforms the existing method, particularly in sparse networks and rough environments with lots of coverage holes.