A Novel Scheme For Contact Predictions In Opportunistic Networks

Abstract

In the opportunistic network (ON) paradigm information is exchanged between two devices as they encounter each other. For such information exchange to take place the devices must know about the presence of other devices in the neighborhood. A very fundamental problem in ON is to predict the occurrence of a future opportunistic contact which is otherwise highly dynamic and unreliable. An accurate predictor which takes into account the long time history can benefit from multiple objectives. Such a predictor switches to the data transfer mode in anticipation of a contact. Also it maximizes the number of opportunistic contacts while spending minimal energy. In this thesis, we have designed a predictive framework and evaluated it using data mining methodologies to accurately predict opportunistic contacts. For evaluation of our scheme, we have used the Bluetooth traces collected by University of Illinois at Urbana Champaign movement (UIM) framework using Google Android phones for a period of 3 weeks. Extensive simulation of our scheme using these real life traces show that the precision and recall values are close to 50% higher compared to the previous schemes. Also the energy usage, is 35% lower for KFP making it an attractive option for predicting opportunistic contacts, to obtain efficient routing as well as swift information dissemination in ONs in an energy efficient manner.