HiPCV: History based learning model for predicting contact volume in Opportunistic Networks

Abstract

In absence of fixed infrastructure in Opportunistic Networks (OppNet), connectivity between OppNet nodes (usually characterized by human-portable devices), is one of the most challenging issues. The traditional assumption considers every proximity triggered human contact to be an effective OppNet connection. However, the high dynamicity of human mobility impairs the interchangeable notion of human contact and effective oppnet connection, thus necessitating the consideration of other critical contact properties like contact volume, defined as the maximum amount of data transferable during a contact. Recently a few works were proposed to predict the contact volume, using the instantaneous movement direction and velocity of the users. However none of those considered previous mobility history of the users which has a significant role on the future estimations. In this paper, we propose a novel scheme called HiPCV, which uses a distributed learning approach to capture preferential movements of the individuals, with spatial contexts and directional information and paves the way for mobility history assisted contact volume prediction. Experimenting on real world human mobility traces, HiPCV first learns and structures human walk patterns, along her frequently chosen trails. By creating a Mobility Markov Chain (MMC) out of this pattern and infusing it into HiPCV algorithm, we then devise a decision model for data transmissions during opportunistic contacts. Experimental results show the robustness of HiPCV in terms mobility prediction, reliable opportunistic data transfers and bandwidth saving, at places where people show regularity in their movements.